MSC Report JoanMasSoldevilla 4947096 FinalVersion

J.
Mas Soldevilla
DYNAMIC ANALYSIS AND STABILITY STUDY OF THE

ELECTROMAGNETIC SUSPENSION LEVITATION SYSTEM
OF THE HYPERLOOP
Hyperloop case
Dynamic analysis and stability study of the
electromagnetic suspension levitation system of
the Hyperloop
Hyperloop case
By
J. Mas Soldevilla
in partial fulfilment of the requirements for the degree of
Master of Science
in Civil Engineering
at the Delft University of Technology,

to be defended publicly on Monday April 28th, 2022 at 10:00 AM.
Student number: 4947096

Project duration: March 15th, 2021 – April 28th, 2022
Thesis committee: Dr. ir. K. N. Van Dalen, TU Delft, supervising chair
Prof. dr. ir. A. V. Metrikine, TU Delft, supervisor
Dr. ir. J. M. de Oliveira Barbosa, TU Delft, weekly supervisor
ir. A. Faragau, TU Delft, bi-weekly supervisor
ir. A. Bougioukos, Witteveen+Bos N. V., weekly supervisor
An electronic version of this thesis is available at http://repository.tudelft.nl/.
ii
iii
Preface
After having a very interesting experience at TU Delft during a one year exchange program at the
Civil Engineering Faculty, I felt very attracted to the course offer presented by the university in the field of
structural engineering and structural dynamics and I decided to take on the challenge of moving to the
Netherlands and take part in such a hurdle MSc programme. Throughout these years, I have learned much
more new crafts and technologies than I would have ever imagined, and I have met many extraordinary
people along the way. It has not been a straight path, but more like a roller coaster. Now, taking a look back,
I feel very proud of all the academic, professional and personal achievements reached over the last two and a
half years. I feel very blessed to have been part of such an incredible academic institution as TU Delft and to
have had the chance of working along side many experts.
As an engineer, I am striving to develop engineering projects and concepts to cope with the current
transportation, social, and environmental challenges by means of innovative, sustainable, and digitalized
technologies and tools, always aiming at improving human welfare. This is what made me decide on taking
the structural engineering MSc path, so as to be able to shape the cities and infrastructures of the future,
trying to make a positive impact in society. This brought me to follow the annotation in structural dynamics,
since it is a craft that is not that much develop in my country of origin. For these reasons, when Karel van
Dalen proposed me to take part in the Hyperloop project, I did not hesitate to accept. I think this technology
is going to be a game changer in the transportation sector as we currently know it and will provide cleaner,
more efficient, more accessible, faster and better integrated transportation between relevant demographic
node and goods distribution hubs.
I feel very grateful by the unique opportunity that Apostolos Bougioukos and Witteveen+Bos gave
me to conduct this thesis with them. It has been an unbelievable opportunity to grow my professional career,
improve my technical skills and my team working aptitude and to learn more about the Dutch culture.
I could not have completed this research without the help and support of all the members of my
committee and I want to express my sincere thanks to all of them. My company supervisor Apostolos
Bougioukos, and my daily supervisors Andrei Faragau and Joao de Oliveira have been the backbone of this
project, working alongside me in every step of the way and making me give my best version. I am also very
grateful to Karel van Dalen, who I first know as my first professor in the university and trusted me all the
way until the date, providing me my first professional opportunity in my internship with Witteveen+Bos and,
afterwards, giving me the opportunity to take part in this project and offering himself to be the chair of my
committee. I feel very fortunate of having had the chance to be your a student and having worked alongside
you all this time. I also want to thank professor Andrei Metrikine for his valuable input in every committee
meeting, always bringing an alternative point view, inviting to critically think about any assumption and
conclusion, and through which I gained very valuable knowledge.
Finally, I want to thank my family and friends for their unbelievable support, in such a hard time
marked by a pandemic. A special thanks goes to my partner and my parents, for always keeping my back,
helping me and cheering me in the very lows, and making me maintain my feet on the ground in the very
highs. I am grateful for all the sacrifices and all the hours you devoted to me, so that I could reach this
moment one day.
J. Mas Soldevilla
Delft, April 2022
iv
v
Abstract
The Hyperloop is a high-speed means of transport, consisting of a bullet-shape vehicle travelling in a
quasi-vacuum tube, electrically powered and moving through an electromagnetic levitating system. This
allows the reduction of air resistance and wheel-rail contact friction, which translates in higher speeds for
smaller power inputs. In the Netherlands, the Dutch company Hardt Hyperloop, has developed a new design
concept of such technology, by means of electromagnetic suspension systems, unlike the most commonly
applied levitating techniques.
The main aim of this project is to investigate the stability of the electromagnetic suspension levitation
system, study the effect that the implementation of an error-based closed-loop control system has on the
system dynamics and investigate the vehicle-structure interaction once a control system has been applied to
the system.
Along these lines, the overall infrastructure-vehicle system is modelled through an equivalent two
degrees of freedom system. In this manner, the inherent instability of the electromagnetic system is
confirmed and the initial vehicle dynamics are highlighted. Afterwards, an error-based closed-loop PD-
control system that applies to the system definition and reacts on the uncontrolled system dynamics is
implemented and the effect of the control parameters on the dynamics and the stability of the system is
studied. A certain stability region in the parametric space is derived, which ensures the stability of the system
for significant perturbations of the vehicle from its equilibrium point. Besides, the existence of a subcritical
bifurcation with respect to the parameter combination is demonstrated. The safety margin in the time delay
of the controller response is studied, so as to define a slack time that accounts for not only processing and
sampling delays, but also unforeseen events. Finally, the vehicle-infrastructure interaction is studied and its
stability is ensured, by means of the previously defined control scheme.
vi
vii
List of Contents
Abstract ............................................................................................................................................................. vi
LIST OF FIGURES ........................................................................................................................................... x
LIST OF TABLES .......................................................................................................................................... xiii
1. INTRODUCTION ......................................................................................................................................... 1
1.1. BACKGROUND .................................................................................................................................... 1
1.2. PROBLEM STATEMENT ..................................................................................................................... 2
1.3. RESEARCH OBJECTIVES ................................................................................................................... 3
1.4. REPORT OUTLINE ............................................................................................................................... 4
2. LITERATURE REVIEW AND THEORETICAL FRAME ......................................................................... 6
2.1. MAGNETIC LEVITATION TECHNIQUES ........................................................................................ 6
2.1.1. Electromagnetic suspension (EMS) ................................................................................................. 7
2.1.2. Electrodynamic suspension (EDS) ................................................................................................... 8
2.1.3. Comparison of different magnetic levitation technologies ............................................................ 10
2.1.4. Stability of magnetic levitation technologies ................................................................................. 10
2.1.5. Research and major achievements in the field ............................................................................... 11
2.2. ELECTROMAGNETICS FUNDAMENTALS .................................................................................... 12
2.3. STRUCTURAL DYNAMICS REVIEW .............................................................................................. 14
2.4. STABILITY .......................................................................................................................................... 15
2.4.1. One dimensional flow .................................................................................................................... 15
2.4.2. Two-dimensional flows.................................................................................................................. 18
2.5. CONTROL ............................................................................................................................................ 26
3. MODELLING OF THE VEHICLE EMS LEVITATION SYSTEM .......................................................... 31
3.1. RAIL INFRASTRUCTURE MODEL .................................................................................................. 31
3.2. EMS AND VEHICLE MODEL ........................................................................................................... 36
4. STABILITY STUDY OF THE OPEN-LOOP EMS LEVITATION SYSTEM.......................................... 42
4.1. VEHICLE DYNAMICS FOR THE OPEN-LOOP EMS LEVITATION SYSTEM ........................... 42
4.2. VEHICLE DYNAMICS FOR THE OPEN-LOOP EMS LEVITATION SYSTEM CONSIDERING
THE VARIABILITY OF THE ELECTRICAL CURRENT ....................................................................... 49
5. STABILITY STUDY OF THE CLOSED-LOOP EMS SYSTEM WITH P-CONTROL........................... 54
6. STABILITY STUDY OF THE CLOSED-LOOP EMS SYSTEM WITH PD-CONTROL ........................ 64
6.1. SYSTEM RESPONSE STUDY BASED ON CONTROL PARAMETERS ....................................... 72
6.2. SYSTEM RESPONSE STUDY BASED ON INITIAL CONDITIONS .............................................. 84
6.3. STUDY APPEARANCE OF A POSSIBLE NUMERICAL ARTIFACT ........................................... 92
7. ADDITION OF A FORCE ON/OFF SWITCH IN THE EMS SYSTEM WITH PD-CONTROL ............. 94
8. STUDY OF THE VEHICLE – TRACK INTERACTION ......................................................................... 99
9. CONTROL DELAY SAFETY MARGIN FOR STABILITY .................................................................. 109
10. CONCLUSIONS AND RECOMMENDATIONS .................................................................................. 117
10.1. CONCLUSIONS ............................................................................................................................... 117
10.2. RECOMMENDATIONS .................................................................................................................. 118
BIBLIOGRAPHY .......................................................................................................................................... 120
ANNEX A: ELECTROMAGNETISM ......................................................................................................... 123
ANNEX B: STRUCTURAL DYNAMICS REVIEW ................................................................................... 131
B.1. SINGLE DEGREE OF FREEDOM ................................................................................................... 131
B.1.1. Free Vibrations ............................................................................................................................ 132
B.1.2. Forced Vibrations ........................................................................................................................ 137
B.1.3. SDOF Stability with respect to Damping .................................................................................... 139
B.2. MULTIPLE DEGREE OF FREEDOM ............................................................................................. 140
B.3. SIMPLY SUPPORTED EULER-BERNOULLI BEAM MODAL VIBRATION ............................ 140
viii
B.4. LAPLACE TRANSFORM ................................................................................................................. 141
ANNEX C: STABILITY ............................................................................................................................... 144
C.1. ONE-DIMENSIONAL FLOW BIFURCATIONS ............................................................................ 144
ix
LIST OF FIGURES
Figure 1.1 Longitudinal cut hyperloop system lay out featuring the levitation system and the vertical motion
DOF .................................................................................................................................................................... 3
Figure 2.1 Representation of the different magnetic levitation systems [18] .................................................... 7
Figure 2.2 EMS system used in the Transrapid [18].......................................................................................... 8
Figure 2.3 Initial EMS system concept projection Hardt Hyperloop [21] ......................................................... 8
Figure 2.4 Pod design XP-2 Test Vehicle Virgin Hyperloop [23] ..................................................................... 9
Figure 2.5 Phase portrait of the system [33] .................................................................................................... 16
Figure 2.6 Circle flow [33]............................................................................................................................... 18
Figure 2.7 Example phase portrait trajectories for system with 1  2 for 2  1  0 [33] ......................... 20
Figure 2.8 Example trajectories behaviour tending to the fixed point and tending to t→ -∞ [33] .................. 20
Figure 2.9 Example phase portrait for a) center and b) spiral [33] .................................................................. 21
Figure 2.10 Example star node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 ≠ 0 [33] .......................................................... 22
Figure 2.11 Example degenerated node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 = 0 [33] ............................................ 22
Figure 2.12 Representation in phase portrait of a) an ordinary node and b) a degenerated node [33] ............ 22
Figure 2.13 Diagram representing the classification of fixed points [33]........................................................ 23
Figure 2.14 Representation of attracting point and Liapunov stability [33] .................................................... 24
Figure 2.15 Limit cycle classification [33] ...................................................................................................... 24
Figure 2.16 Parameter family of closed orbits ................................................................................................. 24
Figure 2.17 Equilibrium decay and supercritical bifurcation growth depending on the control parameter [33]
.......................................................................................................................................................................... 25
Figure 2.18 Behaviour subcritical Hopf bifurcation before and after it happens............................................. 26
Figure 2.19 Closed-loop feedback control general structure [35].................................................................... 26
Figure 2.20 PID Controller closed-loop feedback scheme [36] ....................................................................... 27
Figure 3.1 Hardt Hyperloop infrastructure layout sketch ................................................................................ 32
Figure 3.2 Hardt Hyperloop infrastructure tender European Test Centre [7] .................................................. 32
Figure 3.3 Hardt Hyperloop vehicle run along tube simulation [7] ................................................................. 32
Figure 3.4 Equivalent Euler-Bernoulli, represented under the influence of the electromagnet ....................... 34
Figure 3.5 Overall cross-section model of the Hyperloop vehicle, levitation system and track infrastructure34
Figure 3.6 Drawings of the hollow tube and the rectangular prismatic beam ................................................. 35
Figure 3.7 Beam SDOF oscillator equivalent model ....................................................................................... 36
Figure 3.8 Model of the EMS levitation system cross-section with reference system definition.................... 37
Figure 4.1 Root Locus mapping of the SDOF system in the complex plane ................................................... 46
Figure 4.2 Phase Portrait SDOF vehicle for EMS open-loop levitation system .............................................. 47
Figure 4.3 Phase Portrait with dynamic behaviour classes SDOF vehicle ...................................................... 48
Figure 4.4 Root Locus mapping for1.5DOF open loop system ....................................................................... 52
Figure 5.1 Definition PID-Control scheme for the EMS levitation system stabilization Each of the branches
of control correspond to a component of the PID-control: the proportional P-component, the integral I-
component and the derivative D-control. ......................................................................................................... 55
Figure 5.2 Vehicle position imposing fixed point 1 at  0 ............................................................................... 57
Figure 5.3 Electrical current imposing fixed point 1 at  0 .............................................................................. 57
Figure 5.4 Voltage for target air gap equilibrium position for fixed point 1 at  0 .......................................... 57
Figure 5.5 Vehicle position imposing fixed point 2 at  0 ................................................................................ 58
Figure 5.6 Electrical current imposing fixed point 1 at  0 .............................................................................. 58
Figure 5.7 Voltage for target air gap equilibrium position for fixed point 2 at  0 .......................................... 58
Figure 5.8 Real part of the eigenvalues with respect to the proportional control gain .................................... 61
Figure 5.9 Imaginary part of the eigenvalues with respect to the proportional control gain ........................... 61
x
Figure 5.10 Nonlinear solution of the system with P-control against time ...................................................... 62
Figure 6.1 Eigenvalues at the negative real half-space with respect to K p and K d ....................................... 68
Figure 6.2 1st eigenvalue stability region of negative real part within the parametric plane K p − K d ............ 69
Figure 6.3 2nd eigenvalue stability region of negative real part within the parametric plane K p − K d ........... 69
Figure 6.4 3rd eigenvalue stability region of negative real part within the parametric plane K p − K d ........... 70
Figure 6.5 Intersection pole surfaces with 0-plane. Limits of stability region ................................................ 71
Figure 6.6 Definition of the stability region..................................................................................................... 71
Figure 6.7 Nonlinear and linearized solutions of the system against time for a K p value close to the left
boundary of the stability region ....................................................................................................................... 73
Figure 6.8 Phase portrait of the system for a K p value close to the left boundary of the stability region ..... 74
Figure 6.9 Nonlinear and linearized solutions of the system against time for a K p value centered within the
stability region.................................................................................................................................................. 75
Figure 6.10 Phase portrait of the system for a K p value centered within the stability region ....................... 76
Figure 6.11 Nonlinear and linearized solutions of the system against time for a K p value close to the right
Figure 6.12 Phase portrait of the system for a K p value close to the right boundary of the stability region . 78
Figure 6.13 Plot of the real part of the three eigenvalues of the linearized system with respect to K p .......... 79
Figure 6.14 Nonlinear and linearized solutions of the system against time for a K p value smaller than the
left boundary of the stability region ................................................................................................................. 80
Figure 6.15 Phase portrait of the system for a K p value smaller than the left boundary of the stability region
.......................................................................................................................................................................... 81
Figure 6.16 Nonlinear and linearized solutions of the system against time for a K p value exactly at the right
Figure 6.17 Phase portrait of the system for a K p value exactly at the right boundary of the stability region
.......................................................................................................................................................................... 82
Figure 6.18 Nonlinear and linearized solutions of the system against time for a K p value after the right
Figure 6.19 Phase portrait of the system for a K p value after the right boundary of the stability region ...... 83
Figure 6.20 Nonlinear and linearized solutions for study case within vehicle dynamics region 1 .................. 85
Figure 6.21 Phase portrait of the system for study case within vehicle dynamics region 1 ............................ 86
Figure 6.22 Nonlinear and linearized solutions for study case within vehicle dynamics region 2 and 3 ........ 86
Figure 6.25 Nonlinear and linearized solutions for study case within vehicle dynamics region 4 .................. 88
Figure 6.27 Phase portrait of the system for K p = 20000 and K d = 28000 with all the initial conditions
listed in the legend and represented by a orange circle .................................................................................... 90
listed in the legend and represented by a orange circle .................................................................................... 91
Figure 6.29 Small stable orbit around the fixed point in the phase portrait od the system .............................. 92
Figure 7.1 Nonlinear solution of the closed-loop EMS levitation system with PD-Control ........................... 96
Figure 7.2 Nonlinear solution of the closed-loop EMS levitation system with PD-Control and force on/off
switch ............................................................................................................................................................... 97
Figure 7.3 Phase portrait of the system for K p = 60000 and K d = 100000 .................................................... 98
Figure 8.1 Definition of the stability region................................................................................................... 102
xi
Figure 8.2 Nonlinear solution of the system against time for a K p value close to the left boundary of the
stability region................................................................................................................................................ 104
Figure 8.3 Phase portrait of the system  − & for a K p value close to the left boundary of the stability region
........................................................................................................................................................................ 105
Figure 8.4 Nonlinear solution of the system against time for a K p value centered within the stability region
........................................................................................................................................................................ 105
boundary of the stability region ..................................................................................................................... 106
Figure 8.6 Plot of the real part of the three eigenvalues of the linearized system with respect to K p ........... 107
Figure 9.1 Example of distorting delay in time domain ................................................................................ 110
Figure 9.2 Example of a transport delay in time domain ............................................................................... 110
Figure 9.3 Eigenvalues plotted versus the time delay for K p value close to the left boundary of the stable
region ........... 114
Figure 9.4 Eigenvalues plotted versus the time delay for K p value in the middle of the stable region ....... 115
Figure 9.5 Eigenvalues plotted versus the time delay for K p value close to the right boundary of the stable
region.............................................................................................................................................................. 115
Figure A.1 Magnetic field produced by a moving electric charge [29] ......................................................... 124
Figure A.2 Magnetic fields in solenoid (left) and permanent magnet (right) [29] ........................................ 124
Figure A.3 Representation attraction (a) and repulsion (b) forces [29] ......................................................... 125
Figure A.4 Magnetic field formed around a conductor carrying current [29] ............................................... 126
Figure A.5 Simple magnetic circuit [29]........................................................................................................ 127
Figure A.6 The hysteresis loop traced out by the flux in a core when the alternating current is applied to it: a)
alternating current, and b) hysteresis loop [29] .............................................................................................. 128
Figure A.7 Electromagnetic induction [29] ................................................................................................... 129
Figure A.8 Induced current [29]..................................................................................................................... 129
Figure A.9 Forces on current-carrying conductors in magnetic field [29] .................................................... 130
Figure A.10 Coil's reaction to increasing current [29] ................................................................................... 130
Figure B.1 SDOF mass-spring-dashpot system with F(t) random time-dependent external load ................. 131
Figure B.2 Free Vibrations of the mass-spring system  = 0 [31] ................................................................ 134
Figure B.3 Aperiodic free motion of mass-spring-dashpot system   −1    1 [31] ............................... 135
Figure B.4 Free vibrations of mass-spring-dashpot system −1    1    0 [31] ..................................... 136
Figure C.1 Saddle-node bifurcation example ................................................................................................ 144
Figure C.0.2 Vector field stack [33]............................................................................................................... 145
Figure C.0.3 Saddle-node bifurcation [33] .................................................................................................... 145
Figure C.4 Transcritical bifurcation example ................................................................................................ 145
Figure C.5 Supercritical pitchfork bifurcation vector field [33] .................................................................... 146
Figure C.6 Supercritical pitchfork bifurcation [33] ....................................................................................... 146
Figure C.7 Subcritical pitchfork bifurcation [33] .......................................................................................... 146
xii
LIST OF TABLES
Table 3.1 Input numerical value of model parameters ..................................................................................... 41
Table 6.1 Summary of the study cases presented in this section ..................................................................... 72
Table 6.2 Summary of the study cases presented in this section ..................................................................... 84
Table 8.1 Summary of the study cases presented in this section ................................................................... 103
Table B.1 Operations for Laplace transform [31] .......................................................................................... 142
Table B.2 Table of Laplace transforms [31] .................................................................................................. 142
xiii
xiv
CHAPTER1: INTRODUCTION
1
INTRODUCTION
1.1. BACKGROUND
With the current globalized society, the demand for more time-, power- and capacity-efficient
modern transport is growing. Along these lines, the market share for high-speed transportation is
increasing rapidly and, by 2050, it is predicted that around half of the total traffic in the world will be
high-speed transportation [1]. In the last decades, a significant growth in the amount of high-speed
rail lines has been experienced and, currently, fully connected high-speed rail networks have been
constructed covering the whole Europe, Japan and China, which arise as the corner stone of this
technological development. As observed over the last twenty years, the future of transport will meet
technological advances in electric power, digitalisation and automation [2].
Whilst commercial aviation is still the cheapest and fastest means of transport for long-
distance trips, its efficiency in terms of travel time (including commuting time to the service
infrastructure, check-in, boarding, trip duration, landing, debarking and commuting time to the final
destination), fuel consumption, pollution and frequency is more limiting. In spite of recent efforts of
aircraft manufacturers to design more fuel efficient prototypes, commercial and freight aviation still
represents one of the most polluting industries. In 2018, global aviation (i.e. commercial, freight and
military aviation) represented a 2.5% of the total CO2 emissions worldwide and, with respect to non-
CO2 climate impacts, it accounts for 3.5% of the global warming [3]. This becomes even more
significant when one realizes the small amount of the worldwide population that has access to
aviation. In this manner, the idea of a high-speed, sustainable, environmentally friendly, affordable
and high-frequency traffic alternative was born, aiming at substituting such short-haul aviation
routes.
1
CHAPTER 1: INTRODUCTION
The origin of such an alternative goes back to the 1799, when George Medhurst patented an
atmospheric railway which could transport both cargo and people pneumatically through a
pressurized cast iron tube. This initial design relied on steam power propulsion. This idea was later
refined by Robert H. Goddard in 1904, presenting the so called vactrain, defined as a high-speed rail
transportation moving through evacuated tubes in a reduced air environment with relatively little
power. He wrote down his vision of it in The High-speed Bet, which was published in 1909 under the
name The Limit of Rapid Transit. The main concept was then improved and further developed by
several different engineering, always acknowledging that the biggest limitation for rail transportation
to reach high-speed was the presence of air resistance.
Following the aforementioned ideas, in 2013, a new technology was proposed under the tag
of Hyperloop, as an open-sourced technological concept to be developed in a participative manner
for any company and research institutes in the world, with the goal to make all these ideas become
real. The initial concept consisted of pressurized bullet-shaped capsules riding on air bearings driven
by linear induction motors and axial compressors throughout reduced-pressure tubes [4]. This design
was conceived to cope with friction, leading to faster seeds than conventional rail with significantly
less power. Beside, the design was conceived to be electrically powered, avoiding the polluting
emissions from aviation and road traffic. However, considering economic and engineering
constraints, the initial idea has been modified leading to different designs which may vary
significantly between companies and countries.
In this thesis, the focus is brought on the design proposed by Hardt Hyperloop, the company
developing this technology in the Netherlands, which differentiates significantly from the rest of
current designs by using an electromagnetic suspension system, a very innovative vehicle-
infrastructure lay out and proposing a design compatible both for people and cargo transportation.
This chapter aims to introduce the problem statement, the objectives and the structure of the
research addressed in this thesis project.
1.2. PROBLEM STATEMENT
Unlike other Hyperloop designs, Hardt Hyperloop’s proposal is based on an electromagnetic

suspension system, i.e. it uses electromagnets that attract a magnetically conductive track to maintain
the vehicle at a constant spacing hanging from the rails. This poses an important challenge since, in
general, the electromagnetic suspension is dynamically unstable unless properly controlled [5].
Besides, at the moment, there are no references on the study of electromagnetic suspension levitation
systems applied to Hyperloop technologies. Consequently, this thesis tries to extend the theory in
electromagnetic suspensions system developed up to the date for Maglev technologies to the
Hyperloop design concept.
Hence, this thesis focuses on two things. Firstly, on addressing the challenge of stabilizing
this vertical levitation system, such that the vehicle maintains a specific air gap with the rails at all
times, by means of an error-based closed-loop PID-controlling scheme type. Secondly, on
investigating the effect that this controller can have of the levitation system dynamic, understanding
how the control parameters govern the vehicle dynamics (Fig. 1.1).
2
The sketch below illustrates the problem addressed in this report, representing a longitudinal
cut of the vehicle and the track infrastructure of the Hardt Hyperloop design.
Figure 1.1 Longitudinal cut hyperloop system lay out featuring the levitation system and the vertical motion
DOF
Along these lines, this project is built around: i) the understanding of the stability of the
electromagnetic suspension levitation system, ii) the parametric study of several magnitudes of
interest that take part in this system, iii) the investigation of the application of an error-based closed-
loop control on the system that ensures the target air gap between rail and vehicle proposed by the
company, iv) the analysis of the dynamic interaction between the rail and the vehicle, considering a
flexible guideway modelling.
The design and implementation of electromagnetic suspension systems in Hyperloop entail

interdisciplinary concepts such as electronics, electromagnetism, mechanical engineering, structural
dynamics, control engineering and measurements.
1.3. RESEARCH OBJECTIVES
The stability of Electro-Dynamic Suspension (EDS) applied to Hyperloop technology and the
stability of Electro-Magnetic Suspension (EMS) applied to Maglev technologies have been studied
in several papers [6-18]. However, there are no references on the study of EMS applied to Hyperloop
technologies. Consequently, the first objective addressed in this thesis should be described as
follows: [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [13]
3
1. Extrapolate EMS system definition for the Hardt Hyperloop design.
1.a) Define the formulation and the critical parameters involved.
1.b) Confirm that the system is inherently unstable and explain the reason for such an event.
Once the EMS system stability is better to understood, one can then propose a control
strategy to ensure that the vehicle can be kept stable around an equilibrium target air gap. Along
these lines, the extensive research produced up to date in error-based proportional – (P), derivative-
(D), integral- (I), PD, PI and PID closed-loop controls can be applied for this specific application. As
a consequence, the second objective of this project is described as:
2. Design an error-based closed-loop control following a fundamental approach that can

stabilize the system.
2. a) Define the control variables, the control function and the system plant function.
Generate a clear understanding of the controlling scheme principles.
2. b) Analyse the behaviour of the control concerning its control gains and clearly define a
stability region of the system using a study of the linearized system.
2. c) Provide a simple tuning basics of the control gains for further research steps.
After a clear study of the stability of the EMS system and the tracking of the position of the
vehicle, the research is extended by considering the vibrations of the infrastructure (i.e. beam-rail
and tube). Thus, the third objective of this thesis can be stressed as follows:
3. Study of the vehicle-infrastructure system at the design cross-section, only in the vertical
direction.
1.4. REPORT OUTLINE
This report consists of five main sections: a literature review, a chapter presenting the model
constructed and the derivation of the equation of motions and assumptions, a third chapter presenting
the stability study from the open-loop most simplified system to the model with stabilizing closed-
loop feedback control, another chapter presenting the delay of the linearized system and the dead
time safety margin, and the last chapter featuring the dynamic analysis of the infrastructure-vehicle
system. This is all concluded by a chapter listing the conclusions obtained from this research and the
proposition of further research recommendations.
Chapter 2 serves to present the current state of the art of levitation technologies, their current
applications and the control schemes developed for its stabilization. This section is useful to locate
this project within the current developments and researches on this technology. Besides, its ultimate
purpose is to provide the reader with all the basic engineering concepts that will later face in the
presentation of our research, such that he/she can follow all the developments described.
4
Chapter 3 presents the modelling of the problem addressed in this report and all the
modelling assumptions taken for the derivation of the equations of motion.
Chapter 4 serves to confirm the inherent instability of the EMS levitation system modelled
for this problem and feature the vehicle dynamics without the implementation of any control
technology (i.e. open-loop systems).
Chapter 5 features the modelling and results of the first closed-loop model with the simplest
control scheme existent and how this affects the system response.
Chapter 6 presents the modelling of the closed-loop system applying a PD-control. This
section serves to understand, first of all hoe this control system works and how it affects the vehicle
dynamics, and secondly, to showcase that the stabilization of the system is possible. The sensibility
of the system response with respect to the control parameters is explained and how this gives rise to
certain bifurcations.
Chapter 7 serves to correct the model derived in the previous section by adding an on/off
switch on the electromagnetic force, that avoids misleading electrical current outputs and increases
the stabilization region with respect to the magnitude of the perturbation applied to the system.
Chapter 8 features the stability study of the vehicle-track coupled system and how applying
the error-based PD-control system together with the on/off switch on the electromagnetic force the
system is stable. The sensibility of the system response with respect to the control parameters is
explained and how this gives rise to certain bifurcations.
Chapter 9 presents a study of the effect of time delay on the control system response and hoe
critical it is to keep the system stable.
Chapter 10 summarizes the conclusions obtained from this project and proposes several
recommendations for future researches on the topic.
5
CHAPTER 2: LITERATURE REVIEW
2
LITERATURE REVIEW
AND THEORETICAL
FRAME
Aiming at clarifying concepts that might be used later when modelling the levitation system
and studying the error based closed-loop control implementation, this section will serve to clarify
both electrodynamic suspension systems (EDS) and electromagnetic suspension systems (EMS)
conceptual levitating suspension systems, elaborate on the stability patterns resulting from the study
of the eigenvalues of the system, review relevant theory on the dynamic behaviour of structures, and
shed light on the principles of the control engineering and define several design alternatives.
2.1. MAGNETIC LEVITATION TECHNIQUES
Aiming at increasing the railway’s traffic efficiency, the engineers have identified the friction
at the contact interface between the wheels and the rail as a relevant source of energy loss. In this
manner, over the last century, several engineers have focused on the research of levitating rail
infrastructure design which eliminates this friction and allows the vehicles to reach higher speeds
and provides a larger lifetime of the infrastructure. Along these lines, namely two main levitation
technology designs have been developed: electromagnetic suspension (EMS) and electrodynamic
suspension (EDS). The latter has led to two different design concepts for which one uses
superconducting electromagnets and the other uses an array of permanent magnets. The last is better
6
known as Inductrack. Each of these levitation techniques is schematically represented in the figure
below (Fig. 2.1).
Figure 2.1 Representation of the different magnetic levitation systems [18]
2.1.1. Electromagnetic suspension (EMS)
The EMS system uses the attractive magnetic force of electromagnets placed in the vehicle’s
support frame (e.g. boogie) that draws the vehicle to a magnetically conductive track, maintaining it
at a target spacing from the rails. The electromagnets use feedback control systems to actively
monitor the air gap and maintain the train at a constant distance from a track [19].
There is little literature available on the Hyperloop, especially on the EMS system due to its
reduced application to date. Moreover, the Hyperloop technology itself is at a proof of concept stage.
However, wider research has been devoted to its application in Maglev trains applications, which can
be extrapolated to our design. The electromagnetic levitation system of Maglev trains, has attracted
much attention in recent years, due to its low noise, non-contact operation, and strong climbing
ability [7].
For Maglev trains applications the EMS system can lift a train using attractive forces by the
magnets beneath a guide-rail [15]. The onboard suspension electromagnet (or permanent magnet plus
excitation control coils) that is placed under the track is energized to generate the electromagnetic
field through the mutual attraction between the electromagnet and the rail. This is captured in the
figure below (Fig. 2.2.). The vehicle is suspended on the rail. The suspension gap between
electromagnet and electromagnetic rail is about 8–10 mm [20].
7
Figure 2.2 EMS system used in the Transrapid [18]
This levitation system has only been implemented in one Hyperloop design project, the
Hardt Hyperloop. Nevertheless, this mechanism changes slightly its layout when it comes to Hardt
Hyperloop’s Design. In this case, the rail is placed on the ceiling of the hyperloop tube and consists
of a magnetically conductive track, and the hybrid permanent magnets are placed fixed to the boogie
frames which move below the rail, as shown in the figure below (Fig. 2.3.).
Figure 2.3 Initial EMS system concept projection Hardt Hyperloop [21]
2.1.2. Electrodynamic suspension (EDS)
In the EDS system both the rail and the train exert a magnetic field, and the train is levitated
by the repulsive force between these magnetic fields. The magnetic field in the train can be produced
by either superconducting electromagnets (e.g. JR-Maglev) or by an array of permanent magnets
(e.g. Inductrack).
8
The repulsive force in the track is created by an induced magnetic field in wires or other
conducting strips in the track [19]. However, the EDS train is suspended only when it reaches a
certain speed. When the train is running, the moving magnetic field of the onboard magnet will
excite the induced current in suspension coils (always low-temperature superconducting coils or
permanent magnets) installed in the line. The moving field interacts with the induced current and
produces an upward force to draw the car upon the road surface by a certain height around 10–15 cm
generally [20].
Along these lines, at slow speeds, the current induced in these coils and the resultant
magnetic flux is not large enough to support the weight of the train. For this reason, the train must
have wheels or some other form of landing gear to support the train until it reaches a speed that can
sustain levitation [19]. A slight increase in distance greatly reduces the repulsive force and returns
the vehicle to the right separation again [20], meaning that the gravity is no longer a destabilizing
action as in EMS, but a stabilizing one
This levitation technique is the most widely used on the globe for Hyperloop applications.
Especially significant is its use by the two leading companies in Hyperloop development: Hyperloop
Transportation Technologies (HTT) and Virgin Hyperloop.
The technology has been licensed to HTT from Lawrence Livermore National Labs (LLNL),
which developed it as part of the Inductrack system. This method is thought to be cheaper and safer
than traditional maglev systems. With this method, magnets are placed on the underside of the
capsules in a Halbach array. This focuses the magnetic force of the magnets on one side of the array
while almost entirely cancelling out the field on the other side. These magnetic fields cause the pods
to float as they pass over electromagnetic coils embedded in the track. Thrust from linear motors
propels the pods forward [22].
HTT’s main rival, Hyperloop One from Virgin Hyperloop is also using a passive magnetic
levitation system where pod-side permanent magnets repel a passive track, with the only input
energy coming from the speed of the pod [22]. The design layout of the Hyperloop One is presented
below (Fig. 2.4.)
Figure 2.4 Pod design XP-2 Test Vehicle Virgin Hyperloop [23]
9
2.1.3. Comparison of different magnetic levitation technologies
Each of the aforementioned systems has advantages and disadvantages. A good

understanding of these limitations and positive aspects will allow us to define attention points and
drawbacks to address in the implementation of the EMS in the Hyperloop concept.
The main advantage of EMS systems is that they can work at all speeds and avoid the
disadvantage of EDS which only work at a minimum speed of about 30 km/h, which can eliminate
the requirement for a separate low-speed suspension system and can simplify the track layout as a
result [20]. Its application in recently developed Maglev trains has proven to be a commercially
available technology that can attain very high speeds of up to 500 km/h [19].
In contrast, the main disadvantage of the EMS technology is the dynamic unstable nature of
electromagnetic attraction, explained in detail in the next section, which requires the constant
monitoring of the levitation air gap which is controlled and corrected by computer systems integrated
either on the rail or the vehicle [19]. These control systems can be centralized or decentralized.
On the other hand, the main advantage of EDS systems is that they are naturally stable. A
minor narrowing in distance between the track and the magnets creates the strong forces to repel the
magnets back to their original position. No feedback control is needed. Moreover, powerful onboard
superconducting magnets and a large margin between rail and train enable the highest recorded train
speeds of up to 581 km/h and heavy load capacity [19].
Whereas the main disadvantages are the fact that at the slow speed, the resultant magnetic
flux and the current which is induced in these coils is not large enough to support the weight of the
Maglev train. For this reason, the train must have wheels or some other form of landing gear to
support the train until it reaches a speed that can sustain the levitation. Since a train may stop at any
location, due to equipment problems, for instance, the entire track must be able to support both low-
speed and high-speed operations, which increases significantly the costs [20]. Another disadvantage
is that the EDS system naturally can produce a field in the track in front and to the rear of lift
magnets, which can act against the magnets and produce the drag force [24]. Besides, such strong
magnetic fields onboard the train might make the train inaccessible to passengers with pacemakers or
magnetic data storage media such as hard drives and credit cards [19].
2.1.4. Stability of magnetic levitation technologies
Static magnetic bearings using only electromagnets and permanent magnets are unstable, as
explained by Earnshaw's theorem. Samuel Earnshaw showed that it is not possible to place a
collection of bodies, subject only to electrostatic forces, in such a way that they remain in a stable
equilibrium configuration [5]. This theorem applies to magnetic forces as well. In fact, it is true for
any force that varies with the inverse square of the distance, or any combination of such forces [25].
In this manner, EMS systems rely on active electronic stabilization. Such systems constantly
measure the bearing distance and adjust the electromagnet current accordingly [19]. Many different
control schemes can be applied for such purpose. Extensive coverage of this concept is provided in
section 2.5.
10
Since all EDS systems are generate the electromagnetic force based on electrodynamic
forces, Earnshaw's theorem does not apply to them. This is an important advantage of this type of
magnetic levitation technique.
2.1.5. Research and major achievements in the field
Electromagnetic levitation is not a new thing, but has been studied for long. However, in
recent years, its study in applications such as Maglev trains and high speed trains has increased
significantly. Lately, several papers have brought the spotlight on the study of the levitation stability,
the stabilization of the system using different control schemes and the stability of the vehicle-track
coupled system for flexible guideways.
The study of the stability of EMS Maglev trains running through a flexible guideway has
been widely studied [6 - 18]. In general, all of them present the same design principles for the EMS
levitation system definition, with slight changes in the magnet shape, and the vehicle-infrastructure
fitting and interaction. The most significant differences are come with the type of control system
applied or the guideway modelling.
Following the literature review, the most commonly applied control scheme are error-based
close-loop PD-control on the electrical current or the voltage across the electromagnet [6-14, 15-18].
These are generally formulated by monitoring levitation air gaps, vertical accelerations of the vehicle
or/and electrical current in the circuit. Others implement also the integral components, leading to
error-based close-loop PID-control [26] or PI-control [15]. Others proposed multiple level feedback
control schemes, controlling electrical current, magnetic air gap flux density and air gap, with
complementary frequency filters [27]. Others use more complex scheme to derive a robust controller,
such as Linear Quadratic Gaussian (LQR) regulator or Sliding Mode Control (SMC) based on
Karman filters [28].
The most common modelling of the flexible guideway is by simplifying the track beam to a
single-span simply supported Euler-Bernoulli beam [6 – 12, 14 - 18]. In these cases, a modal analysis
is then conducted for the model of the beam int the system. Then, as more extensively explained in
chapter 3, the vibration frequency corresponding to the maximum amplitude of track beam with
maglev train running on its upper part is distributed within a range, wherein the first-order frequency
of the track beam has the greatest impact [11]. Therefore, only the influence of the first order
bending mode of the track beam is taken into account in most of the papers [6 - 18]. Others model
the track beam as a double span continuous beam [13].
Junxiong Hu et al. [9] analysed the mechanical characteristics of single electromagnet system
and elastic track beam of EMS maglev train and established a five-dimensional dynamics model of
single electromagnet-track beam coupled system with classical PD control strategy adopted for its
levitation system. In this paper, based on the Hurwitz algebraic criterion and the high-dimensional
Hopf bifurcation theory, the stability of the coupled system is analysed, proving the existence of a
subcritical Hopf bifurcation governed by the K p value [9].
11
The investigation time delay in the control system has also been addresses for the EMS
Maglev levitation system of on a flexible guideway, like in the paper written by Junqi Xu et al. [14],
where the study of a the Hopf bifurcation with double time-delay feedback of maglev train running
on the flexible guideway is presented considering time-delayed position feedback signal  1 and
velocity feedback signal  2 . A novel method is presented to develop the double-parametric Hopf
bifurcation diagram in relation to  1 and  2 [14].
R. P. Talukdar et al. [13] outline an approach for the modelling and simulation of Maglev
vehicle–guideway in a block diagram environment, optimizing the suspension parameters for
increased ride comfort. They modelled the guideway as a two span continuous beam and four EMS
suspension frames connecting the total vehicle length with the beam. In this paper, the guideway
surface roughness is defined by power spectral density function. They proved that the vehicle
operating speed is the primary criteria that govern the magnitude of response of both the guideway
and the vehicle. They concluded that the two-span continuous guideway performs better than a
single-span guideway because of the redistribution of sagging moment to the interior support. It is
showed that the guideway irregularity has a greater impact on the car-body response than on the
guideway response [13].
Considering the literature review, the main gap covered by this thesis is the application of an
EMS levitation system in a Hyperloop technology, since to date no papers have been published on
the application of this levitation technique for such a technology concept. In this sense, this thesis
aims at understanding the dynamics of this levitation system, investigating the implementation of an
error-based close-loop PD control on the voltage across the electromagnet coil only monitoring the
levitation air gap. In this manner, this report provides a study of the effect of the control parameters
in the system dynamics, of the appearance of bifurcations or other (in)stability patterns, of the
definition of the needed set up to stabilize the system and of the effect on the system stability of
possible time-delay in the control response.
Aiming at meeting these research benchmarks, several different concepts and engineering
methods learned form the aforementioned literature review will be applied in different section, such
as the Hurwitz algebraic criterion, the definition of classical error-based closed-loop PD-controls, the
study of time delayed control responses, etc.
2.2. ELECTROMAGNETICS FUNDAMENTALS
The author of this report has written a theoretical summary of the most important
fundamental concepts of electromagnetism, that later are used to develop the equations of motion of
the system in chapter 3, based on the H. -S. Han et al. book [29]. This can be found in Annex A.
The phenomenon of magnetic levitation is developed through magnetic fields between

magnetic objects. Such magnetic fields are generated by the movement of electric charges and are
generally represented by field lines with no beginning or end, forming closed loops. The field lines
come from the North pole (N) and enter the South pole (S), they are continuous and do not meet each
other.
12
In this manner, an attractive force is generated between two magnetized objects when the
field lines go from one to another. Whereas if the same poles are facing each other, the field lines are
pressured, resulting in a repulsive force between them.
A flux 𝜙 is a series of field lines. The value of the flux corresponds to the number of field
lines, as a measure of the magnetic field’s strength. Such that, the more lines, the stronger the
magnetic field and the larger the flux. Then, the flux density 𝐵 is defined as the flux per unit area
normal to a magnetic field.
If electrons with negative charges flow through a conductor, a magnetic field is produced
around it. This field is called an electromagnetic field. Magnetic permeability  represents the
relative ease of establishing a magnetic field in a given material. Whereas, reluctance  is a
magnetic resistance in materials, which is the counterpart of electrical resistance.
An electromagnet is defined as a type of magnet in which the magnetic field is produced by

an electric current. Generally, it consists of a large number of closely spaced turns of wire that create
the magnetic field. Along these lines, the magnetomotive force Fm is the cause of the magnetic flux
in a magnetic circuit. Therefore, it can be defined as the effective current flow applied to the core
Fm = NI [ A  N o turns] , where I is the electrical current and N is the number of coil turns around
the core. The degree to which a magnetic field by a current can magnetize a material is called
magnetizing force H , and it is defined as the magnetomotive force Fm per unit length of material.
The magnetic flux B induced in the material depends upon the nature of the material, such that
B = H .
In electromagnetism and electronics, inductance is the property of a conductor by which a

change in current flowing through it induces a voltage or electromotive force in both the conductor
itself (i.e. self-inductance) and any nearby conductors (i.e. mutual inductance). A changing electric
current through a circuit that contains inductance induces a proportional voltage that opposes the
change in current. The relationship among the parameters for a coil with inductance L is defined as
dI
V = IR − L (2.1)
dt
N 2 A
where L = .
l
Considering the aforementioned definitions of electromagnets, it is important to introduce the

concept of Kirchhoff’s law of voltage and current which state that “the total voltage around a loop is
equal to the sum of all the voltage drops within the same loop” and “the total current or charge
entering a junction or node is equal to the charge leaving the node, cause the charge has no other
place to go except to leave as no charge is lost within the node”, respectively. As aforementioned,
this later will allow us to relate the magnetic and electric variables all in one equation such that the
full behaviour of the electromagnet will be captured through the expression.
dI d
V = IR − L −K (2.2)
dt dt
13
Where  describes the levitation air gap. In this manner, eq. 2.11 defines the total voltage of
the system, from its ohmic term input subtracting the associated losses due to variation rates of
electrical current and levitation air gap.
Then, throughout the combination of the magnetomotive force, the magnetizing force,
reluctance and inductance, an electromagnetic force definition can be derived, governing the
electromagnetic suspension levitation system.
2.3. STRUCTURAL DYNAMICS REVIEW

Aiming at modelling the electromagnetic suspension system in Hyperloop, one has to first
address the layout of the mechanism as a simple single degree of freedom mass-spring-dashpot
system considering only the vehicle motion and assuming an infinitely stiff beam. Once the beam
flexibility is accounted for, the problem transforms into a double mass and single spring-dashpot
system.
The vibration of the guideway play an important role in the dynamics of Maglev vehicles.
This is significantly important, considering the fact that the elevated guideway supported by piers
deforms and vibrates when the Maglev vehicle travels over it [8]. Thereby, this yields a vehicle-
structure coupled vibration system. If the parameters of the vehicle system do not match the track
beam system, coupled self-excited vibrations may occur between the electromagnet and the track
beam, affecting the stable levitation of the vehicle [6]. The flexibility of the track beam, which has a
significant impact on the levitation stability ( [16], [30]), is the main cause of the self-excited
vibration of the electromagnet-track beam coupling, especially during stationary or low-speed
operation ( [10], [17]).
The author of this report has written a theoretical summary of the most important
fundamental concepts of structural dynamics on the fields of single degree of freedom (SDOF)
spring-dashpot system dynamics, multiple degrees of freedom (MDOF) system, Euler-Bernoulli
simply supported beam equation, and Laplace transform, that later are used to develop the equations
of motion of the system in chapter 3, based on the H. -S. Han et al. book [29], A. Metrikine structural
dynamics lecture notes book [31] and E. M. J. Vicca MSc thesis [32]. This can be found in Annex B.
Following this summary, the derivation of the equations of motion governing the SDOF
spring-dashpot, both for free and forced vibrations, is presented and system response becomes clear
for different damping values. Besides, from this explanation the influence of the damping ratio on
the system stability becomes obvious. This is then directly applied in the derivation of the equations
of motion of our model, since the definition of SDOF oscillators and a spring-dashpot system play an
important role.
From the explanation presented on Annex B, one can see which parameters play a role when
describing the equation of motion of the track beam that is implemented in the model in a later
section. Especial emphasis is required for the definition of the beam stiffness, structural damping and
mass.
Last, but not least, the summary introduces the concept of the Laplace transform. It is defined
as an integral transform which serves to transform signals and functions from the time domain to the
frequency domain. In this project, this tool is particularly useful, so as to solve linear ordinary
14
differential equations, since in the frequency domain, these become simply algebraic equations. The
Laplace transform of a function G(t ) is defined as
+
t [G (t )] = g ( s ) =  G (t ) exp(− st )dt (2.3)

0
where G(t ) is a function of the real variable t and s is the complex variable, i.e. s =  + i . G(t ) is
called the original function and g ( s) is called the image function [31].
This method will prove to be very useful to compute the poles of our linearized system
around the fixed points, given that the equations of motion are all linear ordinary differential
equations. Besides, it makes the implementation of a control within the system easier to analyse and
formulate in the frequency domain in case a study of the transfer functions, the frequency bandwidth
of stabilization or a phase delay analysis wants to be performed.
2.4. STABILITY
Considering that the main core of this report focuses on the stability study of the EMS
levitation system, it is important to review some basic concepts about stability and sumarize some of
the stability patterns that can be found in the models developed.
There are two main types of dynamical systems: differential equations and iterated maps [33]. This
project only features differential equations, such that the evolution of the system in continuous time
is defined, and its behaviour and stability will feature 1D and 2D flows, limit cycles and bifurcations.
2.4.1. One dimensional flow
The one-dimensional flow can be represented by a dynamic system described as follows
x& = f ( x) (2.4)
Where x(t ) is a real-valued function of time t and f ( x) is a smooth real-valued function of

x . The dependence on t is no explicit, defining an autonomous system as the one derived for this
project. If f would depend explicitly on time, the system would be nonautonomous and,
consequently, two-dimensional.
Then, if one draws the graph of f ( x) and uses it to sketch the vector field on the real line
represented by the x-axis (Fig. 2.5). One can imagine as if a fluid is flowing along the real line with a
local velocity f ( x) . This imaginary fluid is called the phase fluid and the real line is the phase
space. The flow is to the right where f ( x)  0 and to the left where f ( x)  0 . In this manner, to find
a solution to eq. 2.54 starting from an arbitrary initial condition x0 , an imaginary particle called
phase point is placed at x0 and one can then observe how it is carried along by the flow. As time
goes on, the phase point moves along the x-axis according to some function x(t ) . This function is
15
called trajectory based at x0 , and it represents the solution of the differential equation starting from
the initial condition x0 . The representation of one or different trajectories of the system is called a
phase portrait (Fig. 2.5) [33].
Figure 2.5 Phase portrait of the system [33]
The appearance of the phase portrait is controlled by the fixed points x* , defined by
f ( x* ) = 0 and they correspond to stagnation points of the flow. In the example resented in Fig. 2.5,
the solid black dot is stable fixed point , since the flow is toward it in both sides, and the open dot is
an unstable fixed point, since flow is way from it [33].
In terms of the original differential equation, fixed points represent equilibrium solutions,
which are sometimes called steady, constant or rest solutions, since if x = x* initially, then x(t ) = x*
for all time. An equilibrium is defined to be stable if all sufficiently small disturbances away from it
damp out in time. Thus stable equilibria are represented geometrically by stable fixed points.
Conversely, unstable equilibrium, is which disturbances grow in time, are represented by unstable
fixed points [33].
Note that the definition of stable equilibrium is based on small disturbances and certain large
disturbances may fail to decay. In such a case, the fixed point in question is defined to be locally
stable, but not globally stable. For it to be globally stable, it has to be approached from all initial
conditions [33].
Nevertheless, so far, only graphical methods have been discussed to determine in a

qualitative way the stability of fixed points. For a quantitative assessment, an eigenvalue problem
needs to be conducted. However, nonlinear dynamic systems, like the one formulated for this
project, do not have eigenvalues. Consequently, as a first step, the system needs to be linearized
around the fixed point.
Let x* be a fixed point and let  (t ) = x(t ) − x* be a small perturbation away from x* . To see
whether the perturbation grows or decays, a differential equation for  needs to be derived [33].
d
& = ( x − x* ) = x& = f ( x) = f ( x* +  ) (2.5)
dt
Afterwards, Taylor’s expansion can be applied in eq. 2.5.
f ( x* +  ) = f ( x* ) +  f '( x* ) + O( 2 ) (2.6)
16
Where O( 2 ) denotes quadratically small terms in  . Note that f ( x* ) = 0 since x* is a fixed
point.
& =  f '( x* ) + O( 2 ) (2.7)
Now, if f '( x* )  0 , the O( 2 ) terms are so small compared to it that can be neglected. This
of course has some effect on the system solution. If the focus of the stability study is local, the
nonlinear and the linearized system are almost equal. But, if the spotlight is moved far from the fixed
point, the linearized solution can differ significantly from the nonlinear one. The linearized
approximation about x* is then expressed as [33]
&   f '( x* ) (2.8)
This shows that the perturbation  (t ) grows exponentially if f '( x* )  0 and decays if
f '( x* )  0 . If f '( x* ) = 0 , the O( 2 ) terms are not negligible and a nonlinear analysis is needed to
determine stability [33].
The upshot is that the slope f '( x* ) at the fixed point determines its stability. If one looks
back at the earlier examples, it can be seen that the slope was always negative at a stable fixed point.
The importance of the sign of f '( x* ) was clear from our graphical approach; the new feature is that
now, one has a measure of how stable a fixed point is. This is determined by the magnitude of
f '( x* ) . This magnitude plays the role of an exponential growth or decay rate. Its reciprocal
1/ f '( x* ) is a characteristic time scale; it determines the time required for x(t ) to vary significantly
in the neighbourhood of x* [33].
Besides, there exist a theorem that proves the existence and uniqueness of the solution to the
initial value problem. Consider the initial value problem consisting on eq. 2.4 and initial condition
x(0) = x0 . Suppose that f ( x) and f '( x) are continuous on an open interval R of the x-axis and
suppose that x0 is a point in R . Then the initial value problem has a solution x(t ) on some time
interval (− , ) about t = 0 and the solution is unique. This theorem says that if f ( x) is smooth
enough, then solutions exist and are unique. Even so, there is no guarantee that solutions exist
forever, due to the existence of blow-up phenomena, proper from models of combustion and other
runaway processes [33].
Fixed points dominate the dynamics of first-order systems. The only things that can happen
for a vector field on the real line is that all trajectories either approached a fixed point, or diverged to
 . The reason is that trajectories are forced to increase or decrease monotonically, or remain
constant. To put it more geometrically, the phase point never reverses direction [33].
Thus, if a fixed point is regarded as an equilibrium solution, the approach to equilibrium is

always monotonic, overshoot and damped oscillations can never occur in a first-order system. For
the same reason, undamped oscillations are impossible. Hence there are no periodic solutions to eq.
2.4. These general results are fundamentally topological in origin. They reflect the fact that eq. 2.4
corresponds to flow on a line. If you flow monotonically on a line, you will never come back to your
starting place, that is why periodic solutions are impossible [33].
17
The qualitative structure of the flow can change as parameters are varied. In particular, fixed
points can be created or destroyed, or their stability can change. These qualitative changes in the
dynamics are called bifurcations and the parameter values at which they occur are called bifurcation
points. Bifurcations are important scientifically since they provide models of transitions and
instabilities as some control parameter is varied [33]. The most representative one-dimensional flow
bifurcations are the saddle node bifurcation, the transcritical bifurcation and the supercritical
pitchfork bifurcation. Each of this patterns are described in detail in the summary presented in Annex
C, based on the S. H. Strogatz book [33].
But, there are other one-dimensional flows not governed by 2.4, but rather
& = f ( ) (2.9)
Which corresponds to a vector field on the circle, such that  is a point on a circle and & is
the velocity vector at that point. However, this one-dimensional flow differs significantly from the
previous ones for one property: by flowing in one direction, a particle can eventually return to its
starting position (Fig. 2.6). Consequently, periodic solutions become possible, providing the most
basic model of systems that can oscillate. The rest of properties are similar to flows on the line [33].
Figure 2.6 Circle flow [33]
A vector field on the circle is a rule that assigns a unique velocity vector to each point on the
circle. In practice, such vector fields arise when one has a first-order system as eq. 2.9, where f ( )
is a real-valued 2 - periodic function, i.e. f ( + 2 ) = f ( ) for all  , and assume that f ( ) is
smooth enough to guarantee existence and uniqueness of solutions [33].
2.4.2. Two-dimensional flows
In higher-dimensional phase-spaces, trajectories have much more room to manoeuvre and so,
a wider range of dynamical behaviour becomes possible.
As explained for the one-dimensional flow, the nonlinear system does not have eigenvalues. Hence,
either a linear system is considered as
x& = ax + by
(2.10)
y& = cx + dy
18
or a nonlinear system is considered and then the linearized approximation as described in

section 2.4.1 needs to be applied.
Once the linearized system is defined in the time domain, the eigenvalue problem can be
solved aiming at studying the stability of the system around the fixed point. This stability is governed
by the trajectories described by the system on the phase portrait. For the sake of generality, one seeks
trajectories of the form
x(t ) = exp(t ) v (2.11)
Where v  0 is some fixed vector to be determined and 𝜆 is the growth rate, also to be
determined. If such solutions exist, they correspond to exponential motion along the line spanned by
the vector v . To fins the conditions on v and 𝜆, the expression x(t ) = exp(t ) v is substituted in the
linearized system x& = Ax and obtain  exp(t ) v = exp(t ) Av . Then, cancelling the non-zero scalar
factor exp(t ) yields [33].
Av =  v (2.12)
Which says that the desired straight line solutions exist if v is an eigenvector of A with
corresponding eigenvalue 𝜆. This may be called eigensolution. The eigenvalues of a matrix A are
given by the characteristic equation det( A −  I) = 0 , where I is the identity matrix. For a simple
2  2 matrix, as in this study case [33],
 a b
A =  
 (2.13)
 c d 
the characteristic equation becomes
 a− b 
det   =0 (2.14)
 c d −  
Expanding the determinant yields
 2 −  +  = 0 (2.15)
Where  = trace( A) = a + b and  = det( A) = ad − bc .Then, the solution of the quadratic

equation are:
   2 − 4
1,2 = (2.16)
2
The typical situation is for the eigenvalues to be different 1  2 . In this case, a theorem of
linear algebra states that the corresponding eigenvectors v1 and v 2 are linearly independent, and
19
hence span the entire plane. In particular, any initial condition x0 can be written as a linear
combination of eigenvectors as x0 = c1 v1 + c2 v 2 [33].
For real valued eigenvalues, exponential trajectories are drawn in the phase portrait decaying
up to a straight-line trajectory passing through the fixed point. Either such trajectories are straight
along their entire domain or they are straight around the fixed point. In the most common situation,
for the eigenvalues to be different 1  2 (i.e. zero-multiplicity), one can see both eigensolutions
decay exponentially leading to two different trajectories with different direction passing through the
fixed point, as shown in the example phase portrait presented below (Fig. 2.8).
Figure 2.7 Example phase portrait trajectories for system with 1  2 for 2  1  0 [33]
Along these lines, trajectories typically approach the fixed point tangent to the slow
eigendirection, defined as the direction spanned by the eigenvector with the smaller |𝜆|. Whereas, in
backwards time (𝑡 → −∞), the trajectories become parallel to the fast eigendirection, as shown in
the example phase portrait presented below (Fig. 2.9) [33].
Figure 2.8 Example trajectories behaviour tending to the fixed point and tending to t→ -∞ [33]
On the contrary, if the eigen values are complex numbers, the fixed point is a center or a
spiral, which are illustrated in the phase portrait example presented below (Fig. 2.10. a). If the real
part of the complex conjugate pair of eigenvalues is zero, a center is drawn. Whereas if it is non-zero
a spiral is defined (Fig. 2.10. b) [33].
20
Figure 2.9 Example phase portrait for a) center and b) spiral [33]
Note that centers are neutrally stable, since nearby trajectories are neither attracted to nor
repelled from the fixed point. A spiral would occur if the harmonic oscillator was slightly damped.
Then the trajectories would fail to close, because the oscillator loses a bit of energy on each cycle,
since it is a non-conservative system [33].
The aforementioned statements can be proven considering the eigenvalues definition

provided in eq. 2.16, for which complex eigenvalues are obtained when  2 − 4  0 . Hence, these
eigenvalues can be written as 1,2 =   i where

= = Re( )
2 (2.17)
1
= 4 −  2 = Im( )
2
In this manner, assuming   0 provides different eigenvalues (i.e. zero-multiplicity). Then,

the general solution still hold the same shape, but now 𝑐1,2 ∈ ℂ and v ∈ ℂ. Thus, in order to define
x(t ) , linear combinations of exp[(  i )t ] need to be considered, which can be reformulated using
the Euler’s formula as exp(it ) = cos(t ) + i sin(t ) . Thereby, x(t ) can be defined as a combination
of exp( t ) cos(t ) and exp( t )sin(t ) , such that if 𝛼 > 0 growing oscillations are developed, i.e.
unstable spiral, and if 𝛼 < 0 decaying oscillations are developed, i.e. stable spiral. Besides, if 𝛼 = 0
the eigenvalues are purely imaginary and all the solutions are periodic T = 2 /  . Since the
oscillations have a fixed amplitude, the fixed point is a center [33].
In case the eigenvalues are equal 𝜆1 = 𝜆2 = 𝜆 (i.e. multiplicity 1) two possible eigenvector
setups are defined: either there is a unique eigenvector or there are two independent eigenvectors.
When there are two independent eigenvectors, they span the plane and so every vector is an
eigenvector with this same eigenvalue 𝜆. This can be seen by writing an arbitrary vector x0 as a
linear combination of both eigenvectors x0 = c1 v1 + c2 v 2 → Ax0 = A(c1 v1 + c2 v 2 ) = c1 v1 + c2  v 2 =
=  x0 which proves that x0 is also an eigenvector with eigenvalue 𝜆. Since multiplication by A
simply stretches every vector by a factor of 𝜆, the matrix must be a multiple of the identity: If 𝜆 ≠ 0
all trajectories are straight lines through the fixed point x(t ) = exp(t )x0 leading to a star node (Fig.
2.11) ; If 𝜆 = 0 the whole plane is filled with fixed points.
21
Figure 2.10 Example star node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 ≠ 0 [33]
Alternatively, when there is only one eigenvector (i.e. the eigenspace corresponding to 𝜆 in
1D), the fixed point is a degenerated node, as shown in the phase portrait example below (Fig. 2.12),
where as 𝑡 → +∞ and as 𝑡 → −∞ all trajectories become parallel to the only available eigendirection
[33].
Figure 2.11 Example degenerated node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 = 0 [33]
Note that a degenerated node is on the edge between a spiral and a node. The trajectories are
trying to wind around in a spiral, but they do not quite make it. Unlike a degenerated node, an
ordinary node has two independent eigendirections. All trajectories are parallel to the slow
eigendirection close to the fixed point, and to the fast eigendirection as 𝑡 → +∞ (Fig. 2.13) [33].
Figure 2.12 Representation in phase portrait of a) an ordinary node and b) a degenerated node [33]
Considering the variability and complexity of nodes that can be obtain in our stability studies,
a classification of fixed points is proposed below. Such a classification can be conducted namely
using the trace  = 1 + 2 and the determinant  = 12 of the matrix A , which can then be plotted
in a diagram as shown below (Fig. 2.14) [33].
22
Figure 2.13 Diagram representing the classification of fixed points [33]
• If Δ < 0 ⇔ 𝜆1,2 ∈ ℝ ∧ {(𝜆1 < 0, 𝜆2 > 0) ∨ (𝜆1 > 0, 𝜆2 < 0)} → Saddle point
• If Δ > 0 ⇔ 𝜆1,2 ∈ ℝ ∧ {(𝜆1,2 < 0) ∨ (𝜆1,2 > 0)} ∧ 𝜏 2 − 4Δ > 0 → Node
• If Δ > 0 ⇔ {𝜆1,2 ∈ ℂ ∧ {(𝜆1 = 𝜆̅2 ) ∨ (𝜆2 = 𝜆1̅ )}} →
▪ If 𝜏 2 − 4Δ < 0 → Spiral
- If   0 → Unstable spiral
- If   0 → Stable spiral
▪ If 𝜏 = 0 → Centre
• If Δ = 0 ⇔ {(𝜆1 = 0) ∨ (𝜆2 = 0)} → Whole line of fixed points
• If Δ = 0 ⇔ {(𝜆1,2 = 0)} → Plane of fixed points
Where the parabola 𝜏 2 − 4Δ = 0 is the borderline between nodes and spirals. Star nodes and
degenerate nodes lie on this parabola. One can see that centres, stars, degenerate nodes and non-
isolated fixed points are borderline cases that occur along curves on the (Δ, 𝜏) plane, as shown above
(Fig. 2.14) [33].
The stability of nodes and spirals is determined by 𝜏:
▪ For 𝜏 < 0 ⇔ 𝑅𝑒(𝜆1 ) = 𝑅𝑒(𝜆2 ) < 0 → Stable fixed points

▪ For 𝜏 > 0 ⇔ 𝑅𝑒(𝜆1 ) = 𝑅𝑒(𝜆2 ) > 0 → Unstable fixed points
▪ For 𝜏 = 0 ⇔ 𝑅𝑒(𝜆1,2 ) = 0 | 𝜆1,2 = ±𝑖𝜔 → Neutrally stable centres
One can say that 𝑥 ∗ is attracting if there is a 𝛿 > 0 such that lim x(t ) = x* whenever
x→ 
x(0) − x*   , i.e. any trajectory that starts within a distance 𝛿 of 𝑥 ∗ is guaranteed to converge to 𝑥 ∗
eventually (Fig. 2. 15) [33].
Whereas Liapunov stability requires that nearby trajectories remain close for all time as 𝑥 ∗ is
Liapunov stable if for each 𝜀 > 0, there is a 𝛿 > 0 such that x(0) − x*   whenever x(0) − x*  
. Thus, trajectories that start within 𝛿 of 𝑥 ∗ remain within 𝜀 of 𝑥 ∗ for all positive time (Fig. 2. 15)
[33].
23
Consequently, 𝑥 ∗ is asymptotically stable if it is both attracting and Liapunov stable.
Figure 2.14 Representation of attracting point and Liapunov stability [33]
A limit cycle is an isolated closed trajectory, i.e. the neighbouring trajectories are not closed,
but they spiral either towards or away from the limit cycle (Fig. 2.16). If all neighbouring trajectories
approach the limit cycle, the limit cycle is stable or attracting. Otherwise, the limit cycle is unstable,
or in some cases, half-stable [33].
Figure 2.15 Limit cycle classification [33]
Stable limit cycles are very important scientifically, since they model systems exhibit self-
sustained oscillations. In this manner, there is a standard oscillation of some preferred period,
waveform, and amplitude. If the system is perturbed slightly, it always returns to the standard cycle
[33].
Limit cycles are inherently nonlinear phenomena; they can’t occur in linear systems. Of
course, a linear system can have closed orbits, but they won’t be isolated; if x(t ) is a periodic
solution, then so is cx(t ) for any constant c  0 . Hence x(t ) is surrounded by a one-parameter
family of closed orbits (Fig. 2.17). Consequently, the amplitude of a linear oscillation is set entirely
by its initial conditions; any slight disturbance to the amplitude will persist forever. In contrast, limit
cycle oscillations are determined by the structure of the system itself [33].
Figure 2.16 Parameter family of closed orbits
24
The same bifurcation patterns described in section 2.4.1 apply to this case using the same
normal form definitions presented in Annex C for the x equation and adding as the second system
equation y& = − y for all cases. For all these cases the bifurcation occurs when  = 0 , or equivalently,
when one of the eigenvalues is zero, called zero-eigenvalue bifurcations. Such bifurcations always
involve the collision of two or more fixed points. Nevertheless, for the two-dimensional flow, a new
type of bifurcation exists which provides a way for a fixed point to lose stability without colliding
with any other fixed point. These are called Hopf bifurcations. This happens when a pair of complex
conjugate eigenvalues simultaneously crosses the imaginary axis into the right half complex plane.
There are two types of Hopf bifurcation: subcritical and supercritical [33].
Suppose we have a physical system that settles down to equilibrium through exponentially
damped oscillations (Fig. 2.18. a). Now suppose that the decay rate depends on a control parameter
 . If the decay becomes slower and slower and finally changes to growth at a critical value c , the
equilibrium state will lose stability. In many cases the resulting motion is a small-amplitude,
sinusoidal, limit cycle oscillation about the former steady state (Fig. 2.18. b). Then we say that the
system has undergone a supercritical Hopf bifurcation [33].
Figure 2.17 Equilibrium decay and supercritical bifurcation growth depending on the control parameter [33]
In terms of the flow in phase space, a supercritical Hopf bifurcation occurs when a stable
spiral changes into an unstable spiral surrounded by a small, nearly elliptical limit cycle. Hopf
bifurcations can occur in phase spaces of any dimension n ≥ 2. The size of the limit cycle grows
continuously from zero, and increases proportional to  − c , for μ close to c . The frequency of
the limit cycle is given approximately by  = Im( ) , evaluated at  = c . This formula is exact at
the birth of the limit cycle, and correct within O(  − c ) for μ close to c . The period is therefore
T = [2 / Im( )] + O(  − c ) [33].
The subcritical case is always much more dramatic, and potentially dangerous in engineering
applications. After the bifurcation, the trajectories must jump to a distant attractor, which may be a
fixed point, another limit cycle, infinity or, in three and higher dimensions, a chaotic attractor (Fig.
2.19) [33].
25
Figure 2.18 Behaviour subcritical Hopf bifurcation before and after it happens
This sort of bifurcation will be observed in the model showcased in sections 6 and 8 of this
report.
2.5. CONTROL
Aiming at understanding the control engineering that will be needed for the stabilization of
the EMS levitation system, this chapter serves to feature the basic concepts of the process control
and to set the basis for the design of the control later applied in this project.
Process controls are necessary for designing safe and productive plants. A variety of process
controls are used to manipulate processes, however the most simple and often most effective is the
PID controller. The controller attempts to correct the error between a measured process variable and
desired setpoint by calculating the difference and then performing a corrective action to adjust the
process accordingly. A PID controller controls a process through three parameters: Proportional (P),
Integral (I), and Derivative (D). These parameters can be weighted, or tuned, to adjust their effect on
the process [34].
This thesis focuses on a fundamental approach to the control design, shaping it as a closed-
loop feedback control. The main structure of a general closed-loop feedback control is described as
follows (Fig. 2. 20).
Figure 2.19 Closed-loop feedback control general structure [35]
Where one has to define the reference setpoint 𝑟(𝑡), the error 𝑒(𝑡) defined as the difference
between the measured value and the reference setpoint, which activates the controller, the controller
transfer function 𝐺𝑐 (𝑡) which depends on the control type and controller coefficients, the controller
output that is the input of the system 𝑢(𝑡), the plant transfer function 𝐺𝑝 (𝑡) which corresponds to the
26
model formulation, the plant output 𝑦(𝑡) and, in some cases, the sensor transfer function ℎ(𝑡). The
symbolic notation for each component described in this paragraph is presented in Fig. 2.21.
This type of process control is known as error-based control because the actuating signal is
determined from the error between the actual and desired setting. The different types of error-based
controls vary in the mathematical way they translate the error into an actuating signal, the most
common of which are the PID-controllers. Additionally, it is critical to understand feed-forward and
feed-back control before exploring P, I, and D controls [34].
The system over which the control is applied is the so called plant. The main idea of this
engineering concept is the definition of a control variable that is altered and then is inputted in the
system. The output of this system is then the new state/correction of the system. Then a monitored
variable is defined which is measured for each loop and is then compared to a reference value in the
defined error. The system can be defined either with a transfer function, in the frequency domain, or
the state space modelling, in time domain. The most common representation is the transfer function
which effectively defines the system as a relation between the output and the input to the system.
output ( s )
G p (s) = (2.18)
input ( s )
Such a transfer function can be directly related to the stability of the system through root
locus. The poles are the values of the system complex variable s that make the transfer function to
tend to , i.e. the roots of the denominator. The poles are the eigenvalues of the system represented
in the complex plane, which illustrates the stability of the system. The zeros are the values of the
system complex variable s that make the transfer function zero, i.e. the roots of the numerator.
Focusing on the controller definition itself, since the PID is often the most effective, it is the
one that is desired to be implemented in this project. The PID controller consists of three components
which can be represented as branches in the control scheme, as shown below (Fig. 2. 21). Along
these lines, it is important to first understand what is the definition of each of these branches and how
do they effect the system
Figure 2.20 PID Controller closed-loop feedback scheme [36]
The proportional control is a form of feedback control. It is the simplest form of continuous
control that can be used in a closed-looped system. P-only control minimizes the fluctuation in the
process variable, but it does not always bring the system to the desired set point. It provides a faster
response than most other controllers, initially allowing the P-only controller to respond a few
seconds faster. However, as the system becomes more complex (i.e. more complex algorithm) the
response time difference could accumulate, allowing the P-controller to possibly respond even a few
27
seconds faster. Although the P-only controller does offer the advantage of faster response time, it
produces deviation from the set point. This deviation is known as the offset, and it is usually not
desired in a process. The existence of an offset implies that the system could not be maintained at the
desired set point at steady state. It is analogous to the systematic error in a calibration curve, where
there is always a set, constant error that prevents the line from crossing the origin. The offset can be
minimized by combining P-only control with another form of control, such as I- or D- control. It is
important to note, however, that it is impossible to completely eliminate the offset, which is
implicitly included within each equation. P-control linearly correlates the controller output (actuating
signal) to the error (difference between measured signal and set point) [34]
u p (t ) = K p e(t ) + b (2.19)
Where 𝑢𝑃 (𝑡) is the controller output, 𝐾𝑃 is the controller gain, 𝑒(𝑡) is the error and 𝑏 is the
bias. The bias and controller gain are constants specific to each controller. The bias is simply the
controller output when the error is zero [34] and the control gain corresponds to the control activity
related to the given error.
The integral control is a second form of feedback control. It is often used because it is able to
remove any deviations that may exist. Thus, the system returns to both steady state and its original
setting. A negative error will cause the signal to the system to decrease, while a positive error will
cause the signal to increase. However, I-only controllers are much slower in their response time than
P-only controllers because they are dependent on more parameters. If it is essential to have no offset
in the system, then an I-only controller should be used, but it will require a slower response time.
This slower response time can be reduced by combining I-only control with another form, such as P
or PD control. I-only controls are often used when measured variables need to remain within a very
narrow range and require fine-tuning control. I controls affect the system by responding to
accumulated past error. The philosophy behind the integral control is that deviations will be affected
in proportion to the cumulative sum of their magnitude. The key advantage of adding a I-control is
that it will eliminate the offset. The disadvantages are that it can destabilize the controller, and there
is an integrator windup, which increases the time it takes for the controller to make changes [34].
I-control correlates the controller output to the integral of the error. The integral of the error
is taken with respect to time. It is the total error associated over a specified amount of time.
1
u I (t ) =
Ti
 e(t )dt + u (t0 ) (2.20)
Where 𝑢𝐼 (𝑡) is the controller output, 𝑇𝑖 is the integral time, 𝑒(𝑡) is the error and 𝑢(𝑡0 ) is the
controller output before integration. The integral time is the amount of time that it takes for the
controller to change its output by a value equal to the error. The controller output before integration
is equal to either the initial output at time t = 0 , or the controller output at the time one step before
the measurement [34].
Unlike P-only and I-only controls, D-control is a form of feed forward control. D-control
anticipates the process conditions by analyzing the change in error. It functions to minimize the
change of error, thus keeping the system at a consistent setting. The primary benefit of D controllers
is to resist change in the system, the most important of these being oscillations. The control output is
calculated based on the rate of change of the error with time. The larger the rate of the change in
error, the more pronounced the controller response will be [34].
28
Unlike proportional and integral controllers, derivative controllers do not guide the system to
a steady state. Because of this property, D controllers must be coupled with P, I or PI controllers to
properly control the system. D-control correlates the controller output to the derivative of the error.
The derivative of the error is taken with respect to time. It is the change in error associated with
change in time [34].
de(t )
uD (t ) = Td (2.21)
dt
Where 𝑢𝐷 (𝑡) is the controller output, 𝑇𝑑 is the derivative time constant, 𝑑𝑒(𝑡) is the
differential change in error and 𝑑𝑡 is the differential change in time. Mathematically, derivative
control is the opposite of integral control. Although I-only controls exist, D-only controls do not
exist. D-controls measure only the change in error. D-controls do not know where the setpoint is, so
it is usually used in conjunction with another method of control, such as P-only or a PI combination
control. D-control is usually used for processes with rapidly changing process outputs [34].
PD-control is combination of feedforward and feedback control, because it operates on both

the current process conditions and predicted process conditions. In PD-control, the control output is a
linear combination of the error signal and its derivative. PD-control contains the proportional
control’s damping of the fluctuation and the derivative control’s prediction of process error. As
mentioned, PD-control correlates the controller output to the error and the derivative of the error
[34].
de(t )
uPD (t ) = K p (e(t ) + Td ) +U (2.22)
dt
Where 𝑢𝑃𝐷 (𝑡) is the controller output, 𝐾𝑝 is the controller gain, 𝑇𝑑 is the derivative time
constant, 𝑒(𝑡) is the error and U is the initial value of the controller. The equation indicates that the
PD-controller operates like a simplified PID-controller with a zero integral term. Alternatively, the
PD-controller can also be seen as a combination of the P-only and D-only control equations. In this
control, the purpose of the D-only control is to predict the error in order to increase stability of the
closed loop system [34].
Proportional-integral-derivative control is a combination of all three types of control

methods. PID-control is most commonly used because it combines the advantages of each type of
control. This includes a quicker response time because of the P-only control, along with the
decreased/zero offset from the combined derivative and integral controllers. This offset was removed
by additionally using the I-control. The addition of D-control greatly increases the controller's
response when used in combination because it predicts disturbances to the system by measuring the
change in error [34].
PID-control correlates the controller output to the error, integral of the error, and derivative
of the error. Hence, the PID control function in time domain is defined as
t Kp
de(t ) Ki = Ti  K d = K pTd
uPID (t ) = K p e(t ) + K i  (e( )d ) + K d ⎯
⎯⎯⎯ ⎯ ⎯ ⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯
⎯
→
0 dt
t
(2.23)
1 de(t )
uPID (t ) = K p (e(t ) + 
Ti 0
(e( )d ) + Td
dt
)
29
Where 𝑇𝑖 is the integration time, 𝑇𝑑 is the derivative time, 𝐾𝑝 is the proportional gain, 𝐾𝑖 is
the integral gain, 𝐾𝑑 is the derivative gain, 𝑒(𝑡) is the error of the system, 𝑢𝑃𝐼𝐷 (𝑡) is the controller
output and 𝑡 is the time. The gains of the controller have a direct effect on the stability of the system.
For certain values of such coefficients an unstable system can be stabilized.
The proportional term is the only one that affects the equilibrium position of the system. The
derivative term relates to the type of stability of the equilibrium point and the integration term relates
to the smoothness of the solution and the minimization of the error oscillation.
30
CHAPTER 3: EMS SYSTEM MODELLING
3
MODELLING OF THE
VEHICLE EMS
LEVITATION SYSTEM
This section serves as the presentation of the model of EMS suspension developed in this
thesis that constitutes the basis of all analytical study cases. Such a model have been developed
following the design proposed by Hardt Hyperloop. The chapter features the modelling of the rail
infrastructure, the modelling of the vehicle, the modeling of the electromagnets and the derivation of
the main equations of motion.
3.1. RAIL INFRASTRUCTURE MODEL
There are several differences between high speed Maglev applications and Hyperloop, but
one that becomes immediately evident is the type of rail infrastructure. Whilst Maglev’s rails are
fully open sky defined and, even with all power and magnetic equipment, barely consist in a couple
of continuous beams elevated which are generally simply supported by piers. Hyperloop’s
infrastructure is radically opposite, aiming at providing a safe quasi-vacuum environment for the
pods to run. In this manner, the entire Hyperloop track is sheltered within a compartmented tube,
31
which is generally elevated and supported periodically by columns as shown in the sketches below
(Fig. 3.1-3.3), and there are two rails fixed to the ceiling of the tube.
Figure 3.1 Hardt Hyperloop infrastructure layout sketch Figure 3.2 Hardt Hyperloop infrastructure
tender European Test Centre [7]
Figure 3.3 Hardt Hyperloop vehicle run along tube simulation [7]
The tube is steel based and thin walled. The actual Hyperloop track is build by 1 m
rectangular section steel beams, placed one after the other and connected to the ceiling by hanging
metallic connections. Considering these specifications, the track beam itself has been modeled as a
continuous solid rectangular section beam, considering the little spacing between the segments and
that the connecting stiffness between the rail segments and the tube is not governing.
32
After an analysis of the current state of the art and the information provided of the design, it
was concluded that the bending stiffness of the tube itself was governing over other stiffness, such as
the shear stiffness of the tube itself or the bending stiffness of the track beam. In this manner the
track model developed for this thesis is an equivalent Euler-Bernoulli beam with an equivalent mass,
second moment of inertia, stiffness and damping resulting from the combination of both the tube and
the track beam.
While conducting studies on the coupled vibration of the EMS maglev train, Shi et al. found
that the high-order vibration mode can only occur with exceptionally high energy excitation [12].
According to the vibration theory of continuous beam, it can be found that the vibration of higher
modes can occur only when the energy of the excitation is very high [37], [14]. The problem from
the latest test results from Li and others on the Changsha maglev line, it can be seen that for a
definite track beam structure, the vibration frequency corresponding to the maximum amplitude of
track beam with maglev train running on its upper part is distributed within a range, wherein the
first-order frequency of the track beam has the greatest impact [11]. Therefore, only the influence of
the first order bending mode of the track beam is taken into account in the analysis of this paper.
Since the length of the electromagnet is much smaller than the span of the track beam, the
electromagnetic force f m can be considered as a concentrated load acting at position x0 on the track
beam when seeking the limit through the generalized force expression.
In this case, without loss of generality, the following assumptions have been adopted so as to
derive the dynamic equations of motion used in the coming chapters:
1) We are interested in reducing the full system to a two degree of freedom system in order to
focus on the control dynamics. Considering that the cross section height, i.e. diameter, is
much smaller than the length of the tube, the tube is modelled as an Euler Bernoulli beam,
and the parameters governing the stiffness and damping of its equivalent spring-dashpot
system representation in the two degrees of freedom system have been derived according to
this beam theory, assuming that the 1st vibration mode is governing in this system.
2) For the sake of simplicity, although the whole track rail is by definition a continuous beam, in
this project, a simple span between piers is modelled as a simply supported beam.
3) The length of the electromagnet is very small compared to the length of one span of tube.
Consequently, the magnet is considered as a point mass.
Along these lines, one span of track is modeled as a simply supported with equal cross-
section Euler-Bernoulli beam defined following the explanation provided in section 2.3.2 as
presented below (Fig. 3.4). Where z represents the position of the vehicle, w is the beam
deflection, x0 is the position of the magnet over the length of the beam, f m is the electromagnetic
force, M is the mass of the magnet/vehicle, R is the electrical resistance, i is the electrical current
and u is the voltage.
33
Figure 3.4 Equivalent Euler-Bernoulli, represented under the influence of the electromagnet
The overall model lay-out developed based on Hardt Hyperloop’s design is presented below
(Fig. 3.5). This sketch provides cross-sectional view of the design concept, giving a good
representation of the fit of the vehicle on the tube, the main degrees of freedom and the
representation of the levitation system (vertical), although it is not at scale.
Figure 3.5 Overall cross-section model of the Hyperloop vehicle, levitation system and track infrastructure
Below the sketch and dimensions of the tube and the continuous rail are presented (Fig. 3.6).
Where t = 0.018 m , r = 1.75 m , L = 30 m , b = 0.25 m and h = 0.1 m . As indicated in the drawings,
all elements are assumed to be made of S355 steel, consequently the density is taken as
 S 355 = 7800 kg / m3 and the Young’s modulus as E = 210  109 N / m2 .
34
Figure 3.6 Drawings of the hollow tube and the rectangular prismatic beam
Thus, the equivalent mass of the beam per l = 1 m length is obtained as
M eq = 2 M beam + M tube = 2bhl  S 355 +  ((r + t ) 2 − r 2 )l  S 355 = 1941.7 kg (3.1)
The equivalent second moment of inertia is defined as
1 1
I eq = 2( I beam + Abeam d 2 ) + I tube = 2( bh3 + Abeam d 2 ) +  ((r + t ) 4 − r 4 ) = 0.35 m 4 (3.2)
12 4
We consider the most adverse set up with the smallest stiffness, that happens when the
simply supported beam is subject to a point load located at the mid-span, considering that the first
vibration mode is governing as aforementioned in this section pg. 52. Then, the beam is transformed
into an equivalent single degree of freedom oscillator, such that it has no beam and length
attributions but only its stiffness and damping parameters (Fig. 3.7).
In this manner, the equivalent static stiffness of the simply supported beam to a point located
at the centre is
48EI eq
keq = = 1.2892  108 N / m (3.3)
L3
The damping ratio of the rectangular prismatic track beams has been assumed to be of 0.5%.
Consequently, the damping coefficient of the beam can be defined as
cbeam
beam = 0.005 = → cbeam =
2 kbeam M beam L (3.4)
= 2  0.005 7.7778  103  195  30 = 32.2492 N  s / M
The determination of the damping coefficient of the tube requires a first assessment of the
viscous damping associated with an evenly elastic and viscous tube. According to F. Orban an
approximation of the material damping coefficient of any structural member can be calculated as
[38]
35
c = DamV  k h = ctube (3.5)
where Dam is the damping belonging to the maximum stress amplitude, V is the volume of the
structure,  k is the cross-section factor and  h length factor of the member. In this case, for large
diameter continuous steel piping structure Dam  0.03 , the volume of the tube is
V =  ((r + t )2 − r 2 ) L = 5.9681 m3 , the cross-section factor for a tube is  k = 1.27(1 − (t / r )) = 1.2569
and the length factor for a simply supported beam is  h = 1 . Then, the damping coefficient and the
damping ratio of the equivalent Euler-Bernoulli beam can be computed as
ceq = 2cbeam + ctube = 2  32.2492 + 0.2250 = 64.7234 N  s / m (3.6)
ctube DamV  k h
 eq = 2  0.005 + = 2  0.005 + = 0.01 (3.7)
2 ktube M tube L 2 ktube M tube L
Figure 3.7 Beam SDOF oscillator equivalent model
3.2. EMS AND VEHICLE MODEL
If one zooms in on the levitation system within Fig. 3.5, the model build for such system is
represented as shown below (Fig. 3.8).
This model has been constructed following the model assumptions presented in the previous
section 3.1 for the beam definition on the cross-sectional EMS system by an equivalent single degree
of freedom oscillator system.
36
Figure 3.8 Model of the EMS levitation system cross-section with reference system definition
Following the model sketch above, the track beam can be seen as to a single-degree-of-
freedom vibrating body with mass M , which refers to the equivalent mass computed in eq. 3.1,
support stiffness keq and damping ceq and the electromagnet-vehicle system of mass m is levitated
below the track beam. The choice of the reference system has been made for a more intuitive
understanding and a better representation of the results.
The dynamics of the oscillator versus the track beam can also be compared to those of a
pendulum hanging from a beam, for which one can find two equilibrium positions: one for the
oscillator hanging under the rail; and the other for the oscillator mounted on top of the rail. However,
the equilibrium position on top of the rail is physically meaningless since the oscillator is constraint
to move below the rail.
The coupling system has two mechanical degrees of freedom, namely the vertical
displacement of the electromagnet z [m] and the deflection of the rail at the levitation position
w [m] . The electromagnetic force between the electromagnet and the track beam f m [ N ] is a
function of the electromagnet coil current i [ A] and the levitation gap  [m] . The coil current i is
driven by a voltage u [V ] acting across the coil, which has a total resistance of R [Ohm] .
The levitation gap  is defined as the relative displacement between the deflection of the
beam and the position of the oscillator.
 = z−w (3.8)
37
Optionally, irregularities present on the track bottom phase embracing the magnetized area
can be added to the levitation gap  model as a time and longitudinal coordinate dependent function
p ( x, t ) .
The electromagnetic force between the electromagnet and the track beam 𝑓𝑚 can be defined
as shown below. Where the relation between electromagnetic constant parameters has been derived
following the theoretical derivation presented in chapter 2.2 and the nonlinear dependency of the
electromagnetic force on the levitation air gap  and the electrical current i is taken from previous
literature [6 - 17].
0 N 2 Am i 2
f m = sgn( ) (3.9)
4 2
where 0 [T  m / A] is the permeability in vacuum, N is the number of turns of the coil, Am is the
area of the magnetic pole of the electromagnet and sgn( ) refers to the sign of the actual air gap.
The latter, has been added to ensure that the electromagnetic suspension is always working in
attraction, so that in the unrealistic case that the oscillator can move above the rail, the
electromagnetic force is always pointing towards the rail. For the sake of simplicity all constant
0 N 2 Am
parameters are grouped in a generic constant C = .
4
Then the electrical parameters can be related by means of the Kirchhoff’s law of total voltage
following the explanation provided in chapter 2.2. Thus, the control voltage can be related to the coil
current variation and the air gap variation as follows [6 - 15]:
1 i
u = iR + L0i& + k L& = iR + 2C i& + 2C 2 & (3.10)
 
where L0 and k L are defined as follows:
0 N 2 Am 1
L0 =
2  (3.11)
 N Am i
2
kL = 0
2 2
Considering the fact that this is an open loop system, we lack a feedback term for the control
of the voltage through the coil. Therefore, for such a model, the voltage is considered to be a
constant parameter, whereas the electrical current 𝑖 and the actual air gap  are variables.
For its inclusion into the equations of motion, eq 3.16 is rewritten in such a way that the left
hand-side term of the equality contains the derivative of the variable of interest, in this case the
electrical current i , for a better fit in the construction of a system of Ordinary Differential Equations
(ODEs). This equation reads
 i z−w i
i& = (u − iR − 2C 2 &) → i& = [u − iR − 2C ( z& − w& )] (3.12)
2C  2C ( z − w) 2
38
The translational degrees of freedom of interest can then be defined by means of the
equilibrium of forces in their respective bodies based on Newton’s second Law and the derivation of
the SDOF equations of motion presented in chapter 2.3. In this manner, two coupled equations of
motion are obtained, written in the same form as eq. 3.18. The first equation governs the vehicle
motion and the second equation governs the track beam motion.
C i2
z = g − sgn( )
&&
m2
(3.13)
C i2
&& = g + sgn( )
w − 2 eqn w& − n 2 w
M 2
where n = keq / M is the natural frequency of the track beam,  eq = ceq / 2 keq M is the damping
ratio of the track beam, and the zero points of z and w are located in the static equilibrium position.
Then, combining all the aforementioned expressions eq. 3.18 and eq. 3.19, an ODE system of
equations can be build formulating all equations of motion governing the system described by our
model (Fig. 3.7). The equations can be converted into an alternative form, by transforming
 x1 x2 x3 x4 x5  =  z z& w w& i and expanding  the system equations is defined as:
x&1 = x2
C x5 2
x&2 = g − sgn( x1 − x3 )
m ( x1 − x3 ) 2
x&3 = x4 (3.14)
C x5 2
x&4 = g + sgn( x1 − x3 ) − 2 eqn x4 − n 2 x3
M ( x1 − x3 ) 2
x1 − x3 x5
x&5 = [u − x5 R − 2C ( x2 − x4 )]
2C ( x1 − x3 ) 2
The system is defined to have two translational degrees of freedom ( x1 for the vehicle and x3
for the beam) and one electronic degree of freedom ( x5 for the electrical current on the
electromagnet). x2 and x4 are only state variables of x1 and x3 , respectively, namely their first
derivative.
It is assumed that the system at t = 0 is defined at its static equilibrium, i.e. the resultant of
the forces on the vehicle vanishes f m − mg = 0 . In order to account for this set up, two parameters
need to be defined accordingly at t = 0 . Namely, that the initial air gap is set to be the target air gap
and the initial voltage input to the system is such that the generated electromagnetic force is exactly
the same as the weight.
39
 | t =0 =  0
(3.15)
u | t =0 = u0 = {u | f m − mg = 0}
Such a voltage input value can be computed as follows
u0
i2 i | t =0 =   | t =0 = 0 u0 2 mg mg
C = mg ⎯
⎯⎯⎯
⎯⎯⎯
R
⎯
⎯⎯⎯
⎯⎯⎯
→ = → u0 =  0 R (3.16)
2 R 0
2 2
C C
Following this definition and the requirements for the system to be in such equilibrium
position, the system set up at t = 0 is expressed as
z0 = w0 +  | t =0 → x1s = x3s −  0
z&0 = 0 → x2s = 0
mg mg
w0 = → x3s = (3.17)
keq keq
w& 0 = 0 → x4s = 0
u0 mg mg
i0 = = 0 → x5s =  0
R C C
After a quick look on the system of equations of motion eq. 3.20 and the electromagnetic
force definition eq. 3.15, one can easily see that the EMS hyperloop levitation system is nonlinear,
which entails some challenges when it comes to solving it. Consequently, in this report, the focus
will be brought on the local stability of the coupled system around the equilibrium point. A constant
in this project is the linearization of the nonlinear systems around the fixed points, which will
provide as with nonlinear solutions and linearized solutions of the system. Given the fact that the
nonlinear system does not have eigenvalues, the linearized system will result very useful to perform
any eigenvalue problem.
Throughout the project, the different problems have been studied both by deriving simplified
analytical models, analytical linearized solutions, and numerical nonlinear simulations through
MATLAB. All the parameters defining the model described above have been accurately defined
numerical. All input numerical values are collected in the table below (Tab. 3.1).
40
Table 3.1 Input numerical value of model parameters
PARMETER VALUE UNTS

Total mass of vehicle 30000 [kg ]
Mass of one Magnet 150 [kg ]
Lumped Mass Oscillator 15300 [kg ]
Gravity 9.81 [m / s 2 ]
Resistance 9.71 [Ohm]
o
N of coil turns 800 [-]
Permeability of air 4  10 −7
[T  m / A]
Magnetized Area 0.25 [m2 ]
Equilibrium voltage 251.7479 [V ]
Target Air Gap 0.015 [ m]
41
CHAPTER 4: STABILITY STUDY OPEN-LOOP EMS LEVITATION SYSTEM
4
STABILITY STUDY OF
THE OPEN-LOOP EMS
LEVITATION SYSTEM
In this section the stability of the initially proposed model for the EMS levitation system of
the vehicle presented in chapter 3.2 will be studied in depth. As explained in chapter 2, following
Earnshaw’s theorem, all EMS systems are unstable. Therefore, the first objective of this section is to
confirm that the derived system is unstable, and second, understand the resulting stability pattern and
how does this translate in the behaviour of the vehicle run.
4.1. VEHICLE DYNAMICS FOR THE OPEN-LOOP EMS

LEVITATION SYSTEM
The voltage across the coil represents the input of energy to the system. If no feedback
control over the voltage is applied, this translates in the fact that the power provided to the system is
constant along time and does not change when the vehicle deviates from its target position. Along
these lines, one can conclude that for an open-loop design the influence of the electronic parameters
on the motion of the vehicle is limited. Consequently, aiming at understanding the vehicle dynamic
42
behaviour, we focus only on studying the first two equations of the ODE system eq. 3.14,
considering a constant electrical current and an infinitely stiff beam.
Since the two equations of motion governing the motion of the vehicle and the beam are
coupled, if one wants to focus on studying only the vehicle motion, this can be done by assuming an
infinitely stiff track beam, which yields
x3 = x4 = 0 (4.1)
The influence of the electrical parameters on the vehicle dynamics is tried to be reduced to
the minimum. This translates in
u0
x&5 = 0  x5 = x5s = (4.2)
R
This assumptions directly affect the definition of the air gap and the electromagnetic force
which become
 = x1
x5s 2 (4.3)
f m = sgn( ) C
2
Considering all the equations presented above, the equations of motion reduce to eq. 4.4,
which has been written based on the standard form used to define the ODE system in eq. 3.14. In this
case, the system state variables are  x1 x2  =  z z& .
x&1 = x2
2
C x5s (4.4)
x&2 = g − sgn( x1 )
m x12
As aforementioned in chapter 3.2, the focus is put on the local stability of the coupled system
around the equilibrium point. Firstly, we determine the system’s equilibrium points, also called fixed
points, are determined.
Following the theoretical explanation in chapter 2.4, the equilibrium points x* are the points
at which the derivative of the state variables is zero (  x& =  x&1 x&2  = f x* = 0 ). From the nonlinear
system eq. 4.4, the equilibrium points of the system can be defined as ( x1* , x2* ) =
{( x1 , x2 ) | x&1 = 0; x&2 = 0} which yields four possible solutions, which need to be sorted out,
considering that for this system, all parameters, namely 0 , Am , N , m, g , u0 , R , are all strictly positive.
43
x&1 = 0 → x2 = 0
sgn( x1 ) gmC
x&2 = 0 → x1 = x5s ⎯
⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯
2 possible solutions
⎯
→
gm
gmC (4.5)
x11 = x5s {x11 
⎯⎯ ⎯⎯ ⎯⎯ ⎯ ⎯⎯ ⎯ ⎯ ⎯
physically possible
→ x*  R }
⎯
gm
− gmC
x12 = x5s { x12 
⎯⎯ ⎯⎯ ⎯⎯ ⎯ ⎯⎯ ⎯⎯ ⎯ ⎯ ⎯
physically not possible
→ x*  R }
⎯
gm
Keep in mind that any equilibrium point must be real-valued by definition. For this reason,
only one fixed point is feasible, as shown in eq. 4.5.
gmC
( x1* , x2* ) = ( x5s , 0) (4.6)
gm
Then, if one inputs the numerical value of the system parameters presented in Tab. 3.1, one
can see that the fixed point represents exactly the initial equilibrium position of the system, described
by the initial conditions. It is noted that, for such a set up, the oscillator is positioned exactly at the
target air gap and the vertical variation of the oscillator position is 0, i.e. ( x1* , x2* ) = (0.015, 0) .
Once the fixed point is defined, the system can be linearized around the fixed point. As a starting
point, a generic two ODE nonlinear system is considered as presented below.
x&1 = f ( x1 , x2 )
(4.7)
x&2 = g ( x1 , x2 )
Let ( x1* , x2* ) be a fixed point. Then, defining two generic variable u , v such that
u = x1 − x1* ; v = x2 − x2* (4.8)
Which denote the components of a small disturbance from the fixed point. In order to see
whether the disturbance grows or decays, the differential equations for u and v need to be derived,
which for example for u reads
f
u& = x& ⎯
⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯
by substitution
⎯
→ u& = f ( x1* + u , x2* + v ) ⎯
⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯
Taylor series
⎯⎯
→ u& = f ( x1* , x2* ) + u | * * +
 x1 ( x1 , x2 )
(4.9)
f sin ce f ( x1* , x2* ) = 0 f f
+v |( x * , x * ) +O(u 2 , v 2 , uv) ⎯
⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯⎯
→ u& = u |( x * , x * ) + v |( x * , x * ) +O(u 2 , v 2 , uv)
 x2 1 2
 x1 1 2
 x2 1 2
Where O(u 2 , v 2 , uv) denotes the quadratic terms in u and v of the Taylor series expansion.
The same procedure can be derived for v& yielding
g g
v& = u |( x * , x * ) + v |( x * , x * ) +O (u 2 , v 2 , uv) (4.10)
 x1 1 2
 x2 1 2
44
Hence, the entire disturbance (u, v) evolves according to
 f  f 

 u&   x  x2   u
   1   + O(u 2 , v 2 , uv) + ...
  =    E5555555555F (4.11)
 v&   g  g   v 

 x  x2 
Higher order terms
 1
( x1* , x2* )
Where the first matrix is defined as the Jacobian of the system evaluated at the fixed point
* *
( x , x2 ) , which is the multivariable analog of the derivative f '( x1* ) .
1
 f  f 

 x  x2 
 1
J =  (4.12)
 g  g 

 x  x2 
 1
( x1* , x2* )
The same procedure can be analogously applied to the system being studied eq. 4.4.
However, since the effect of the higher order terms is generally very small compered to the first
order term and it has been assumed that they can be neglected, leading to the linearized system
presented below. This has consequences that need to be considered. Whilst, locally around the fixed
point the linearized solution will be very close to the real solution, for further regions this solution
will lose accuracy and might not be reliable anymore. The focus is put on the local stability analysis
around the fixed point. This is especially useful considering the little range of tolerance of relative
movement of the vehicle with respect to the beam.
x& = Jx + f (x) (4.13)
Where f (x) is the right hand-side vector of external forces, which in our case is assumed to
be 0.
Once the system has been linearized, the Jacobian matrix of the system can be written as in
eq. 4.14 and we can evaluate it at the fixed point presented in eq. 4.6.
 0 1  0 1
 
   
J =  2C x5s 2   J |( x * , x * ) =  2mg 2  (4.14)
 0  0
 mx3  x gmC
1 2
 1   5s 
Hence, the linearized system around the fixed point reads
 0 1
 x&1     x1 
  =  2mg   
2
(4.15)
 x&2   0  x2 
 x5s gmC 
45
Eq. 4.15 can be equivalently written in a single equation with variable x1 and coefficients
corresponding to the Jacobian terms, which are all strictly positive defined.
x1 − J 21 x1 = 0 where J 21  0
&& (4.16)
In this manner, one can see that the system is governed by a negative stiffness, which implies
that the system is unstable. This can be proven in a clearer way, by transforming the system from the
time domain to the Laplace domain, as defined in chapter 2.3.4, assuming trivial initial condition and
analysing the response of the system to a unit pulse input forcing term.
x1 − J 21 x1 =  (t ) ⎯
&& ⎯⎯ ⎯⎯ ⎯ ⎯ ⎯ ⎯⎯
Laplace trans .
⎯
→ s 2 xˆ1 − J 21 xˆ1 = 1 (4.17)
Then, following the theory presented in chapter 2.4, one can compute the poles of the system,
i.e. the eigenvalues, by deriving the transfer function of x̂1 and finding the roots of its denominator.
1
xˆ1 = ⎯
⎯⎯ ⎯ ⎯
Poles
⎯⎯
→ s 2 − J 21 = 0 → s1,2 =  J 21 =  36.1663 (4.18)
s − J 21
2
This result can be represented in the complex plane as a root locus mapping as follows (Fig.
4.1)
Figure 4.1 Root Locus mapping of the SDOF system in the complex plane
From fig. 4.1, it becomes obvious that the system is unstable, since one of the two
eigenvalues has a negative real part. The poles are represented by a cross. This plot and the
46
numerical value of the eigenvalues have been obtained by applying the numerical value proposed for
the system parameters in Tab. 3.1.
This serves as a confirmation that the system is unstable, which is a conclusion that meets
Earnshaw’s theorem. After eq. 4.18, one can see that both eigenvalues are different from each other,
both are real-valued, they are equal in absolute value but one is positive and the other one is
negative. Following the theory presented in chapter 2.4, this fixed point can be classified as a saddle
point.
Besides, for a fixed point to be stable, both main directions, i.e. eigenvectors, point toward
the node. This is not the case for a saddle point as shown in the plot of the phase portrait of the
system presented below (Fig. 4.2 and 4.3), where one eigen direction points towards the fixed point
and the other moves away.
Considering the system definition eq. 4.4, the eigenvalues of the system have been found to
s1,2 =  36.1663 and the eigenvectors v1 =  −1 −36.1663 and v 2 =  −1 36.1663 ,
T T
be
respectively. This allows to draw a phase portrait based on the position of the vehicle and its vertical
position variation x1 − x2 around fixed point eq. 4.13 of the system as presented below (Fig. 4.2 and
4.3).
Figure 4.2 Phase Portrait SDOF vehicle for EMS open-loop levitation system
Figure 4.2 presents the phase portrait for several random trajectories each for a different
randomly chosen initial condition. The most interesting information that one can derive from such a
phase portrait is a clear picture of the vehicle dynamics behaviour, which can be classified in 4
different behaviour classes, corresponding to different regions of this phase portrait. For this reason,
a second mapping of the results has been produced, aiming at providing a deeper understanding (Fig.
4.3).
47
Figure 4.3 Phase Portrait with dynamic behaviour classes SDOF vehicle
The best way to analyse this plot is to think of the vertical axis ( x1 = 0 ) as the hyperloop
track and the fixed point represents the equilibrium position of the vehicle at Hardt Hyperloop’s
target air gap, below the rail. In this sense, one can look at this figure in a more natural way by
rotating it 90o to the right.
Region N o 1 is governed by vehicle self-weight taking over the electromagnetic force at

some point, which impedes that for any case with initial conditions within this region the vehicle
never reaches to be attracted within the target air gap range. From the moment that the self-weight
force takes over, the vehicle ends up falling down and moving away progressively from the track.
Normally, this behaviour is characteristic for initial condition combinations quite far below the track,
with both upwards (negative) and downwards (positive) velocities, that are not sufficiently high to
lead the vehicle towards this target air gap band. This phenomena is easy to understand given the fact
that the electromagnetic force as defined in eq. 4.3, has the air gap as its denominator squared. Along
these lines, the farther away the vehicle is located, the bigger this denominator and the lower the
electromagnetic force.
Region N o 2 is governed by the electromagnetic force being unable to compensate the

downward forces of the vehicle, lead by high initial downwards velocities. This region is
characterized by vehicles starting located very close to the track, within the target airgap
electromagnetic influence band, but its positive (downwards) velocity combined with the self-weight
load prove to be too much for the electromagnetic force and the vehicle ends up falling down and
moving away progressively from the track. This phenomena follows form the electromagnetic force
definition in the same way that is described for region N o 1 .
Region N o 3 can be seen as quite the opposite from region N o 2 . It is governed by the
electromagnetic force which combined with high upwards vertical velocities lead to a force build up
that pushes the vehicle upwards until it crashes against the rail and gets stuck to it. In this case, self-
weight forces are unable to counteract the attraction force combined with negative position rates
and,, the vehicle cannot be maintained at a position around the target air gap. This behaviour can be
48
explained by reflecting on the electromagnetic force mathematical definition, for which the closer
the vehicle moves to the rail, the shorter the levitation air gap and the larger the electromagnetic
force.
Region N o 4 is then characterized by system set ups for which, generally, the vehicle both
starts and ends up colliding with the rail. This region comprises not only a very small region below
the rale very close to the track beam, for which the air gap is so short, that for any downward
velocity and self/weight the large electromagnetic force cannot be compensated. But also, all
unrealistic set ups for which the vehicle initial conditions are defines such that it is located above the
rail. In this case all loads acting on the vehicle have down ward component (self-weight as always
and electromagnetic force since it is attractive). Thus, even for large negative velocity inputs, the
vehicle ends up falling on the rail. However, this case is not representative for this project, since our
the design does only consider the vehicle levitating below the rail.
In fact, this plot shows the system behaviour around the unstable equilibrium point, at the
target air gap that we aim to, given any possible initial conditions, confirming all the results
presented above.
It can be concluded that this section confirms that the EMS levitation system without control
is inherently unstable due to the electromagnetic force definition. As illustrated with the explanation
of the different mechanical behaviour classes, the squared of a distance magnitude variable in the
denominator is governing these patterns, corroborating that Earnshaw’s law applies on our system.
4.2. VEHICLE DYNAMICS FOR THE OPEN-LOOP EMS

LEVITATION SYSTEM CONSIDERING THE VARIABILITY OF THE
ELECTRICAL CURRENT
This section aims at understanding the role that the electric variables and the Kirchhoff’s law
of total voltage can have in the vehicle stability, when no feedback loop is defined to update the
power input on the system if required. The assumption of having an infinitely stiff track beam eq. 4.1
still holds for this study case.
However, unlike the previous study case, now, the electrical current x5 is no longer assumed
to be constant. However, since the system being studied is open-loop, the input voltage across the
electromagnetic coil is assumed constant and no iterative correction is available. Along these lines,
the voltage is assumed to the constant initial value associated with the target air gap position,
likewise in the previous case.
In this manner, the equations of motion eq. 3.14 are adapted here for the aforementioned
assumptions. The main difference with eq. 4.3 is that the electrical current is now represented by the
time dependent variable i .
49
 = x1
i2 (4.19)
f m = sgn( ) C
2
Considering all the equations presented above, the equations of motion reduce to eq. 4.22,
which has been written based on the standard form used to define the ODE system in eq. 3.15.
Where the equation based on Kirchhoff’s total voltage law derived in chapter 2.2 has been added. In
this case, the system state variables are  x1 x2 x5  =  z z& i .
x&1 = x2
C x5 2
x&2 = g − sgn( x1 ) (4.20)
m x12
x1 x
x&5 =
(u0 − x5 R − 2C 52 x2 )
2C x1
The first thing to be addressed is to define the equilibrium points. Considering the shape of
the system, four fixed points are expected, but in this case with 3 components each, defining the
system in a 3D space, although the system is still effectively a single degree of freedom system with
a complementary variable accounting for the electronics of the system. Thereby, this can be referred
to as a one and a half degree of freedom system.
The equilibrium points of the system can be defined as ( x1* , x2* , x5* ) =
= {( x1 , x2 , x5 ) | x&1 = 0; x&2 = 0; x&5 = 0} which yields four possible solutions, which need to be sorted
out, considering that for this system, all parameters, namely 0 , Am , N , m, g , u0 , R , are all strictly
positive.
x&1 = 0 → x2 = 0
sgn( x1 ) gmC
x&2 = 0 → x1 = x5s ⎯
⎯⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯ ⎯
2 possible solutions
⎯
→
gm
gmC
x11 = x5s {x11 
⎯⎯ ⎯⎯ ⎯⎯ ⎯ ⎯⎯ ⎯ ⎯ ⎯
physically possible
→ x*  R }
⎯ (4.21)
gm
− gmC
x12 = x5s { x12 ⎯⎯ ⎯⎯ ⎯⎯ ⎯ ⎯⎯ ⎯⎯ ⎯ ⎯ ⎯
physically not possible
→ x*  R }
⎯
gm
x x u
x&5 = 0 → 1 (u0 − x5 R − 2C 52 x2 ) = 0 → x5 = 0
2C x1 R
As explained in eq. 4.23, only one fixed point is find to be feasible
gmC u0 u
( x1* , x2* , x5* ) = ( , 0, 0 ) (4.22)
gm R R
As expected, the same fixed point as before has been obtained, corresponding to the initial
equilibrium position at the target air gap, with an equilibrium electrical current equal to the initial
50
equilibrium value, which evaluated using the numerical value of the system parameters presented in
Tab. 3.1 yields ( x1* , x2* , x5* ) = (0.015, 0, 25.92) .
The Jacobian matrix of the system can be written as in eq. 4.23 and we can evaluate it at the
fixed point presented in eq. 4.22.
 
   
   
   
 0 1 0   0 1 0 
   
 2C x5 2   2
2 g  (4.23)
J =  −
2C x5  J | =  2mg −
0   0 
   x5 gmC x5s 
* * *
m x13 m x12 ( x1 , x2 , x5 )
   s 
 1   
 xx xx
(− x5 R − 2C 2 25 ) + 2 2 25 −
x5 x1 x2 
(− R − 2C 2 )  gmC u0 
  −
gm
−
 2C x1 x1 x1 2C x1   0 
 gmC 2 gmC 
 
 
 
 
 x&1   0 1 0   x1 
     
   2mg 2    (4.24)
=  
2g
 x&2  0 −   x2 
   x5 gmC x5s   
   s   
 x&5 
    
 x5 

 gmC u0 
 −
gm
−
 0 
 gmC 2 gmC 
Then, the same method used in the previous chapter can be used to reduce the two first rows
of the matrix system to an equivalently single equation with variable x1 and x5 , and coefficients
corresponding to the Jacobian terms, which are all strictly positive defined, and a second equation
corresponding to the last row with variables x1 and x5 .
x1 − J 21 x1 + J 23 x5 = 0
&&
(4.25)
x&5 + J 32 x&1 + J 33 x5 = 0
In this manner, one can see that the one term in the equations of motion system has a negative
stiffness in the first equation, which implies that the system is unstable. Then, following the
definition in chapter 2.3.4, the system can be transformed into the Laplace domain, assuming trivial
initial condition and analysing the response of the system to a unit pulse input forcing term as
follows.
x1 − J 21 x1 + J 23 x5 = 0
&& ⎯
⎯⎯ ⎯⎯ ⎯ ⎯ ⎯ ⎯⎯
Laplace trans .
⎯
→ s 2 xˆ1 − J 21 xˆ1 + J 23 xˆ5 = 0
(4.26)
x&5 + J 32 x&1 + J 33 x5 =  (t ) sxˆ5 + J 32 sxˆ1 + J 33 xˆ5 = 1
51
1 − J 32 sxˆ1 1 − J 32 sxˆ1
xˆ5 =  s 2 xˆ1 − J 21 xˆ1 + J 23 ( ) = 0→
s + J 33 s + J 33
J 23
−
s + J 33 1
xˆ1 = =− ⎯
⎯⎯ ⎯ ⎯
Poles
⎯⎯
→ (4.27)
J J s 1
s − J 21 −
2 23 32
[ s + J 33 s − ( J 23 J 32 + J 21 ) s − J 21 J 33 ]
3 2
s + J 33 J 23
1 3
[ s + J 33 s 2 − ( J 23 J 32 + J 21 ) s − J 21 J 33 ] = 0 → s1 = − 51.5155; s2 = −0.7243; s3 = 50.7910
J 23
Figure 4.4 Root Locus mapping for1.5DOF open loop system
The root locus is presented in Fig. 4.4, where it becomes obvious that the system is unstable,
since one of the three eigenvalues has a positive real part, i.e. it is represented in the RHS complex
plane. The system is defined by three real-valued eigenvalues two with negative real part and one
with positive real part. These poles correspond to the crosses in the map, whereas the circles
represent the zeros of the system. This plot has been obtained by applying the numerical value
proposed for the system parameters in Tab. 3.1.
This serves as a verification that the system is unstable, even after the inclusion of a third
equation governed by Kirchhoff’s law to the system. Now, it is important to understand which type
of 3D stability pattern governs the dynamic behaviour of the system.
52
Following the theory presented in chapter 2.4, one can see that around the only fixed point a
saddle point of index 1 is defined since all 3 eigenvalues are strictly real, one of them has a positive
real part and the other two have a negative real part. Note that the index indicates the dimensions of
the unstable manifold, 1 dimensional in this case.
In this case, the construction of a 3D phase portrait of the system does not bring any new
information compared to the one presented for the previous study case (Fig. 4.2). Since the system
drown presents very similar eigen directions as the previous one, with an extra eigenvector
associated to the inclusion of Kirchhoff’s law equation which is close to vertical, the projection of
the phase portrait on the phase plane x1 − x2 draws the same behaviour as presented in Fig. 4.2 and
4.3. For this reason, it has not been included to this report, avoiding trivial information.
It can be concluded that this section shows that, even after including the electric variables in
the system, the system is inherently unstable due to the electromagnetic force definition. As
illustrated with the explanation of the different mechanical behaviour classes, the squared of a
distance magnitude variable in the denominator is governing these instability patterns, corroborating
that Earnshaw’s law applies on our system.
53
CHAPTER 5: STABILITY STUDY CLOSED-LOOP EMS LEVITATION SYSTEM WITH
FEEDBACK P-CONTROL
5
STABILITY STUDY OF THE
CLOSED-LOOP EMS
SYSTEM WITH P-CONTROL
In the previous section, the modelled EMS levitating systems have been proven to be
unstable. Hence, a feedback control system is required to make the system stable. The idea behind
this concept is the definition of a tool that monitors continuously the levitation air gap throughout
time and adjusts the voltage input to the EMS system to ensure the stability of the system. In this
manner, the electrical current present in the numerator of the electromagnetic force, redefines the
magnitude of electromagnetic attractive force to compensate such position of the vehicle.
The use of a PID-control seems convenient due to its simplicity, ease of implementation and
robustness. The PID-control is a combination of the PI- (proportional-integral) and PD-
(proportional-derivative) controls. The PID-control is used when the system performance requires
improvement in both the transient and steady-state performance.
One of the main goals of this thesis, as presented in chapter 1.3, is the design of a feedback
control following a pragmatic approach, so as to stabilize the system, analyse the effect of such a tool
on the vehicle-track stability and understand the effect that each components has. The objective is to
build a similar scheme to a PID-control. However, for a better understanding, the overall design
concept has been broken down into simpler components, studying its effect on the system stability.
54
FEEDBACK P-CONTROL
The reference PID-control scheme can be described as follows (Fig. 5.1). The system will
monitor uniquely the levitation air gap  . Then, the target air gap  0 is taken as reference, so that
the definition of the error is e =  −  0 . The control function is applied on the voltage u , which
becomes the input to the system plant. Finally, the outcome of the system is the entire state variable
vector x =  x1 x2 x3 x4 x5  , through which a new  is defined referring to the new air gap that
resulted form the new voltage compensation input to the system activated by an error different than
0.
Figure 5.1 Definition PID-Control scheme for the EMS levitation system stabilization Each of the branches of
control correspond to a component of the PID-control: the proportional P-component, the integral I-
component and the derivative D-control.
This chapter serves to feature the design of the simplest possible control scheme for this
system. A P-control monitoring the levitation air gap and controlling the voltage across the
electromagnetic coil is described. Following the theory presented in chapter 2.5, a P-control scheme
is a form of feedback control that minimizes the fluctuation in the process variable and provides the
faster response among controllers. P-control linearly correlates the controller output (actuating
signal) to the error (difference between measured signal and set point).
In the same way as the model developed in chapter 4, this study case aims at analysing only
the vehicle dynamics. For this reason, the assumption of having an infinitely stiff track beam eq. 4.1
In this manner, the error defined for this analysis is expressed as e = x1 −  0 and the system
outputs  x1 x2 x5  =  z z& i . The monitored variable is the air gap  and the controlled
variable is the voltage u . The controlled is defined in time domain as part of the equations of
motion. The control loop can be described as
u = K p ( x1 −  0 ) + u0 (5.1)
where K p is the proportional control gain to be tuned and u0 is the reference value of the control
value. In this case it is defined as the voltage input that initially defines the system at the target air
gap equilibrium position.
Hence, the system of equations of motion can be defined as
55
FEEDBACK P-CONTROL
x&1 = x2
C x5 2
x&2 = g − sgn( x1 ) (5.2)
m x12
x1 x
x&5 = ( K p ( x1 −  0 ) + u0 − x5 R − 2C 52 x2 )
2C x1
For this reason, the same procedure described in chapter 4.1 and 4.2 will be applied for this
case study. Thus, the first thing to be addressed is to define the equilibrium points. The equilibrium
points of the system can be defined as ( x1* , x2* , x5* ) = = {( x1 , x2 , x5 ) | x&1 = 0; x&2 = 0; x&5 = 0} which
yields four possible solutions, which need to be sorted out. However, in this case study, not all
parameters are all strictly positive. The majority are, namely 0 , Am , N , m, g , u0 , R , are all strictly
positive. But, the proportional control gain K p can be both positively and negatively defined.
Consequently, the identification of the feasible fixed points, is not that straight forward and further
assessment will be required.
From the computation of the equilibrium points, several solutions are obtained, from which
only two are feasible, considering the fact that any equilibrium point must be real-valued by
definition. These are described as
(CK p − gmC )
K p[ ]( K p 0 − u0 ) − K p 0 + u0
(CK p − gmC )( K p 0 − u0 ) − R 2 gm + CK p 2
(x , x , x ) = (
1* 1* 1*
, 0, )
− R 2 gm + CK p 2
1 2 5
R
(5.3)
(CK p + gmC )
K p[ ]( K p 0 − u0 ) − K p 0 + u0
(CK p + gmC )( K p 0 − u0 ) − R 2 gm + CK p 2
(x , x , x ) = (
2* 2* 2*
, 0, )
− R 2 gm + CK p 2
1 2 5
R
In order to determine which fixed point is feasible, first, the voltage and the electrical current
for the system to have a fixed point at the target air gap  0 for both equilibrium points
( Rgm − K p gmC ) R 0 ( Rgm − K p gmC ) 0

u01 = ; x51s =
CK p − R gmC CK p − R gmC
(5.4)
( Rgm + K p gmC ) R 0 ( Rgm + K p gmC ) 0
u02 = ; x52s =
CK p − R gmC CK p − R gmC
Afterwards, if one considers the results obtained for the open loop case featured in chapter 4,
only one fixed point was feasible corresponding to the target air gap position. Along this lines, it is
reflected that the feasibility of both points can be assessed by imposing them to be at x1 =  0 . This
has been done in two steps, first taking the 1st fixed point solution to be at  0 and the 2nd is defined
accordingly to this set up, which yields the following results (Fig. 5.2 - 5.4). In these plots, the
solution (z/cu/u0) _star 11 holds for the 1st fixed point and the (z/cu/u0) _star 21 for the 2nd fixed
56
FEEDBACK P-CONTROL
point. In these plots the two fixed points are represented with respect to K p for the case of imposing
x1 =  0 in fixed point 1.
Figure 5.2 Vehicle position imposing fixed point Figure 5.3 Electrical current imposing fixed point
1 at  0 1 at  0
Figure 5.4 Voltage for target air gap equilibrium position for fixed point 1 at  0
57
FEEDBACK P-CONTROL
Whereas, if the 2nd fixed point is imposed to be at  0 and the 1st is defined accordingly to this
set up, yields the following results (Fig. 5.5 - 5.7). In this plots, the solutions (z/cu/u0) _star 12 holds
for the 1st fixed point and the (z/cu/u0) _star 22 for the 2nd fixed point. In these plots the two fixed
points are represented with respect to K p for the case of imposing x1 =  0 in fixed point 2.
Figure 5.5 Vehicle position imposing fixed point Figure 5.6 Electrical current imposing fixed point
2 at  0 1 at  0
Figure 5.7 Voltage for target air gap equilibrium position for fixed point 2 at  0
From the figures presented above, one can see that there are two admissible fixed points, One
can freely choose the value of u0 such that one of the of the two fixed points lies at x1 =  0 . If the 1st
fixed point is chosen to be at x1 =  0 , then both the electrical current and the voltage at this fixed
58
FEEDBACK P-CONTROL
point will be mostly negative. Whilst, if the 2nd fixed point is chosen to be at x1 =  0 , then both the
electrical current and the voltage at this fixed point will be mostly positive. This means that
irrespective of our choice of u0 , there are two fixed points on the positive and negative half-spaces
(i.e. in the 3D parameter space) of the electrical current x5 .
Since, by definition of our system, we work with positive values of u and i , the second fixed
point is chosen to represented the feasible fixed point. This can be reasoned due to the attractive
EMS system definition. In this manner, focusing on the physical meaning of the system and it
components, a negative electrical current x5 , would translate on a change of polarization of the
electromagnet which would be understood to be working on repulsion instead of in attraction, which
makes completely o sense for such application. Focusing on the voltage u0 similarly, a negative
value would implement a change on the direction of the electrical flow on the circuit. Along these
lines, the 2nd fixed point is defined to be the physically feasible one at x1 =  0 .
Following the fixed point analysis, the eigenvalue problem is also conducted as a function of
K p , since it is the control gain, which is the tunable value that is assumed to make the system stable.
Therefore, one expects to be able to define different regions of stability depending on the value of
this parameter.
The Jacobian matrix of the system can be written as in eq. 5.5 and we can evaluate it at the
fixed point presented in eq. 5.4.
 
 
 
 0 1 0 
 
 2 
J =  2C x5
−
2C x5 
0 
 m x1 3
m x1 2

 
 1 
 [u + K ( x −  ) − x R − 2C x2 x5 + x ( K + 4C x2 x5 )] − x5 x1 (− R − 2C x2 )
 2 
 2C 0 p 1 0 5
x12 1 p
x13
x1 2C x1 
(5.5)
 
 
 
 
 0 1 0 
 
 2C ( Rgm + K gmC ) 2 (− R 2 gm + CK 2 ) + − + 2 

2
2 C ( Rgm K gmC )( R gm CK p 
)
J |( x * , x * , x * ) =  − 
p p p
0
  − + 3
 − + 2 
 m ( K u )( CK R gmC ) m ( K u )( CK R gmC ) 
1 2 5
p 0 0 p p 0 0 p
 
 ( K p 0 − u0 )(CK p + R gmC ) K p − Rgm − K p gmC ( K p 0 − u0 )(CK p + R gmC ) R 
 
 − 
 2C (− R gm + CK p )
2 2
CK p + R gmC 2C (− R gm + CK p )
2 2 
 
59
FEEDBACK P-CONTROL
 
 
 
 
 0 1 0 
 x&1     x1 
   2C ( Rgm + K gmC ) 2 (− R 2 gm + CK 2 ) 2C ( Rgm + K p gmC )(− R gm + CK p ) 2   
  
2
  (5.6)
 x&2  =  −   x2 
p p
0
    − + 3
m( K p 0 − u0 )(CK p + R gmC ) 2   
 x&   m ( K p 0 u 0 )( CK p R gmC )  x 
 5   
  5
 ( K p 0 − u0 )(CK p + R gmC ) K p − Rgm − K p gmC ( K p 0 − u0 )(CK p + R gmC ) R 
 − 
 2C (− R gm + CK p )
2 2
CK p + R gmC 2C (− R 2 gm + CK p 2 ) 
 
Then, the same method used in the previous chapter can be used to reduce the two first rows
of the matrix system to an equivalently single equation with variable x1 and x5 , and coefficients
corresponding to the Jacobian terms and a second equation corresponding to the last row with
variables x1 and x5 .
x1 − J 21 x1 + J 23 x5 = 0
&&
(5.7)
x&5 − J 31 x1 − J 32 x&1 + J 33 x5 = 0
In this case, no conclusions can be drawn about stability of the system solely from the sign of
the coefficients/stiffness od the system, the Jacobian terms cannot be guaranteed to be strictly
positive due to the presence of K p . The sign of the real part of the eigenvalues needs to be
investigated. Along these lines, following the same method previously applied in chapter 4, the
system in eq. 5.7 in the Laplace domain reads
x1 − J 21 x1 + J 23 x5 = 0
&& ⎯
⎯⎯ ⎯⎯ ⎯ ⎯ ⎯ ⎯⎯
Laplace trans .
⎯
→ s 2 xˆ1 − J 21 xˆ1 + J 23 xˆ5 = 0
(5.8)
x&5 − J 31 x1 − J 32 x&1 + J 33 x5 =  (t ) sxˆ5 − J 31 xˆ1 − J 32 sxˆ1 + J 33 xˆ5 = 1
1 + J 31 xˆ1 + J 32 sxˆ1 1 + J 31 xˆ1 + J 32 sxˆ1

xˆ5 =  s 2 xˆ1 − J 21 xˆ1 + J 23 ( ) = 0→
s + J 33 s + J 33
J 23
−
s + J 33 1
xˆ1 = =− (5.9)
J ( J + J 32 s ) 1 3
s 2 − J 21 + 23 31 [ s + J 33 s 2 + ( J 23 J 32 − J 21 ) s − J 21 J 33 + J 23 J 31 ]
s + J 33 J 23
1 3
⎯
⎯⎯ ⎯ ⎯
Poles
⎯⎯
→ [ s + J 33 s 2 + ( J 23 J 32 − J 21 ) s − J 21 J 33 + J 23 J 31 ] = 0
J 23
The poles of the expression in eq. 5.11 have then been evaluated with respect to K p , which
yield to the following patterns (Fig. 5.8 and 5.9). These are presented in two plots, the first one
featuring the real part of the eigenvalues Re{s1,2,3 } and the second one featuring the imaginary part
Im{s1,2,3 } .
60
FEEDBACK P-CONTROL
Figure 5.8 Real part of the eigenvalues with respect to the proportional control gain
Figure 5.9 Imaginary part of the eigenvalues with respect to the proportional control gain
61
FEEDBACK P-CONTROL
From the figures above, one can see namely three regions. On the one hand, there are two
semi-finite regions of the K p for which two of the eigenvalues are complex conjugates: in the LHS
plane poles green and red, and in the RHS plane poles blue and red; and one of the eigenvalues
strictly real: in the LHS plane the blue eigenvalue, and in the RHS plane the green one. It is
important to realize that when the real part of the complex poles is negative, the real part of the real
eigenvalue is positive, and vice versa. The distribution of those depends on the sign of the
proportional gain. As a consequence, one can conclude that for K p values within this regions, the
system is unstable.
On the other hand, there is a third region between the two aforementioned for which the three
eigenvalues are strictly real. Similarly to the previous cases, when two of the poles are positive (blue
and red), the other one is negative (green), and when two of the eigenvalues are negative (green and
red), the other one is positive (blue). Consequently, one can conclude that for K p values within this
regions, the system is unstable.
Figure 5.10 Nonlinear solution of the system with P-control against time
This becomes clear when plotting the position of the vehicle and the electromagnetic force
against time using the nonlinear simulation. From this figures, it becomes obvious that the control
takes over and, unlike in the previous section for which the vehicle moves progressively away, in
this case the motion of the vehicle is reversed, the electromagnetic forced is increase and the vehicle
is pushed upwards until it clashes with the track. Along these lines, two conclusions can be drown:
first, the control has an effect on the vehicle motion; second, as proven with the eigenvalue problem,
the system is still unstable, but as shown in this case, it either falls down or crashes with the track.
62
FEEDBACK P-CONTROL
In this manner, it is proven that for such a proportional control the system is still unstable and
there is no possible value of K p for which it can be stabilized. Hence, a higher level of complexity
need to be applied to the control scheme definition.
63
FEEDBACK PD-CONTROL
6
STABILITY STUDY OF THE
CLOSED-LOOP EMS
SYSTEM WITH PD-
CONTROL
Despite of the implementation of an initial feedback control system in the one and a half dof
system closed loop system, the stabilization of the system has not been reached. Aiming at designing
a control system like the one described in in chapter 5 and presented in Fig. 5.1, in this section, a D-
control component is added to the previously derived P-control scheme, increasing the control
system by adding another control gain and another control variable.
Following the theory presented in chapter 2.5, the derivative control (D-control) is a feed
forward control that anticipates the process conditions by analysing and minimizing the change in
error. D-control correlates the controller output to the derivative of the error. The derivative of the
error is taken with respect to time.
In the same way as the model developed in chapter 4 and 5, this study case focuses only on
the vehicle dynamics. For this reason, the assumption of having an infinitely stiff track beam eq. 4.1
64
FEEDBACK PD-CONTROL
Following the design trend of the error-based closed-loop control presented in chapter 5, for
this control the error defined for this analysis is expressed as e = x1 −  0 and the system outputs
 x1 x2 x5  =  z z& i . The monitored variable is the air gap  and the controlled variable is the
voltage u . The controlled is defined in time domain as part of the equations of motion. Then, the
control loop can be described as
 ( x1 −  0 )
u = K p ( x1 −  0 ) + K d + u0 = K p ( x1 −  0 ) + K d x2 + u0 (6.1)
t
Where K p is the proportional control gain and K d is the derivative control gain, both to be
tuned, and u0 is called the bias, which namely is the reference value of the control value. In this case
it is defined as the voltage input that initially defines the system at the target air gap equilibrium
position.
Hence, the system of equations of motion can be defined as
x&1 = x2
C x5 2
x&2 = g − sgn( x1 ) (6.2)
m x12
x1 x
x&5 = ( K p ( x1 −  0 ) + K d x2 + u0 − x5 R − 2C 52 x2 )
2C x1
As aforementioned in chapter 3.2, the focus is put on the local stability of the coupled system
around the equilibrium point. Then, the equilibrium points of the system can be defined as
( x1* , x2* , x5* ) = = {( x1 , x2 , x5 ) | x&1 = 0; x&2 = 0; x&5 = 0} which yields four possible solutions, which need
to be sorted out. However, in this case study, not all parameters are all strictly positive. The majority
are, namely 0 , Am , N , m, g , u0 , R , are all strictly positive. But, the proportional control gain K p and
the derivative control gain K d can be both positively and negatively defined. Consequently, now, the
identification of the feasible fixed points, is not that straight forward and further assessment will be
required.
From the computation of the equilibrium points, several solutions are obtained, from which
only two are feasible, considering the fact that any equilibrium point must be real-valued by
definition. These are described as
(CK p − gmC )
K p[ ]( K p 0 − u0 ) − K p 0 + u0
(CK p − gmC )( K p 0 − u0 ) − R 2 gm + CK p 2
(x , x , x ) = (
1* 1* 1*
, 0, )
− R 2 gm + CK p 2
1 2 5
R
(6.3)
(CK p + gmC )
K p[ ]( K p 0 − u0 ) − K p 0 + u0
(CK p + gmC )( K p 0 − u0 ) − R 2 gm + CK p 2
(x , x , x ) = (
2* 2* 2*
, 0, )
− R 2 gm + CK p 2
1 2 5
R
65
FEEDBACK PD-CONTROL
If one compares the components of the feasible fixed points obtained in this case with the ones
previously derived, they are identical. This can be explained by the fact that K p is the only control
parameter which conditions the location of the fixed points of the system. Whereas K d relates to the
stability nature of the fixed points.
Considering that the two feasible fixed points are the same as in chapter 5, the same exact
procedure of analysing the feasibility of both alternatives is applied here, which yields the same
result. The only admissible fixed point is the second one. The reasoning is the same, the definition of
a negative electrical current and voltage is not correct in this design.
Now, the eigenvalue problem is not only conducted as a function of K p , but also as a
function of K d since both are control gains, which are the tunable value that is assumed to make the
system stable. Therefore, one expects to be able to define different regions of stability depending on
the value of these parameters, but now in a 3D space, representing the eigenvalues as surfaces.
Once the fixed point has been calculated, following eq. 4.9 and 4.11, the next step is to define
the Jacobian matrix of the system and evaluate it at the fixed point. However, since the expressions
get more and more involved as the complexity of the system increase, several terms have been
defined as complementary parameters in order to shorten the expressions. These are
 = (CK p + R gmC ) ,  = (− R 2 gm + CK p 2 ) ,  = (− RK d K p 0 − R 2 gm + CK p 2 + RK d uo )
 = ( K p 0 − u0 ) ,  = ( Rgm + K p gmC ) ,  = (−2CRgm − CK d u0 ) .
 
 
 
 0 1 0 
 
 2 
J =  2C x5
−
2C x5 
0 
 m x1 3
m x1 2

 
 1 x2 
 x2 x5 x2 x5
[u + K p ( x1 −  0 ) + K d x2 − x5 R − 2C 2 + x1 ( K p + 4C 3 )]
1
( K d x1 −
2Cx5 x1
(− R − 2C 2 )
 )
 2C 0 x1 x1 2C x1 2C x1  (6.4)
 
 
 
 0 1 0 

 
 2C 2 2C 
J |( x * , x * , x * ) =  0 − 
 m 3 m 2 

1 2 5
 
  K p −2 K p gmC + CK d K p 3 0 +  K p 2 + K d K p R 2 gm 0 + 2 R 3 g 2 m 2 − K d R 2 gmu0  R 
 −
 
 2C 2 2C 
66
FEEDBACK PD-CONTROL
 
 
 
 0 1 0 
 x&1     x1 
    
   2C  2C   
2
 (6.5)
 x&2  =  0 −  x2 
   m 3
 x&   m 2   
 x 
 5    5
 − 
 2C 2 2C 
Then, the same trick used in the previous chapter can be used to reduce the two first rows of
the matrix system to an equivalently single equation with variable x1 and x5 , and coefficients
variables x1 and x5 .
x1 − J 21 x1 + J 23 x5 = 0
&&
(6.6)
x&5 − J 31 x1 − J 32 x&1 + J 33 x5 = 0
In the same manner as chapter 5, no conclusions can be drawn about stability of the system
solely from the sign of the coefficients/stiffness od the system, the Jacobian terms cannot be
guaranteed to be strictly positive due to the presence of K p and K d . The sign of the real part of the
eigenvalues needs to be investigated. Along these lines, following the same method previously
applied in chapter 4 and 5, the system in eq. 6.6 in the Laplace domain reads
x1 − J 21 x1 + J 23 x5 = 0
&& ⎯
⎯⎯ ⎯⎯ ⎯ ⎯ ⎯ ⎯⎯
Laplace trans .
⎯
→ s 2 xˆ1 − J 21 xˆ1 + J 23 xˆ5 = 0
(6.7)
x&5 − J 31 x1 − J 32 x&1 + J 33 x5 =  (t ) sxˆ5 − J 31 xˆ1 − J 32 sxˆ1 + J 33 xˆ5 = 1
1 + J 31 xˆ1 + J 32 sxˆ1 1 + J 31 xˆ1 + J 32 sxˆ1

xˆ5 =  s 2 xˆ1 − J 21 xˆ1 + J 23 ( ) = 0→
s + J 33 s + J 33
J 23
−
s + J 33 1
xˆ1 = =− (6.8)
J ( J + J 32 s ) 1 3
s 2 − J 21 + 23 31 [ s + J 33 s 2 + ( J 23 J 32 − J 21 ) s − J 21 J 33 + J 23 J 31 ]
s + J 33 J 23
1 3
⎯
⎯⎯ ⎯ ⎯
Poles
⎯⎯
→ [ s + J 33 s 2 + ( J 23 J 32 − J 21 ) s − J 21 J 33 + J 23 J 31 ] = 0
J 23
The poles of the expression in eq. 6.11 have then been evaluated with respect to K p and K d ,
which yield to the following patterns (Fig. 6.1 – 6.4). These are presented in four plots. The first one
features the negative real half-space with respect to K p and K d with the three eigenvalues Re{s1,2,3 }
(Fig. 6.1). The reason for this follows from stability theory as explained in chapter 2.4, for which the
stability of any system is, in general terms, ensured by the unique and sufficient condition that all
eigenvalues have negative real part.
67
FEEDBACK PD-CONTROL
Figure 6.1 Eigenvalues at the negative real half-space with respect to K p and K d
From this plot, it can be seen that there is actually one region for which, namely the positive
quarte in the K p and K d plane, where the three eigenvalues coincide, taking a negative real part.
This can be seen as a stable region in the parametric space, for which combinations of gain values
when tuning the control, provide a stabilization of the system. With the objective of analysing this
region, the next three plots feature the stability region of negative real part eigenvalues within the
parametric plane K p − K d for each of the three eigenvalues (Fig. 6.2 – 6.4).
68
FEEDBACK PD-CONTROL
Figure 6.2 1st eigenvalue stability region of negative real part within the parametric plane K p − K d
Figure 6.3 2nd eigenvalue stability region of negative real part within the parametric plane K p − K d
69
FEEDBACK PD-CONTROL
Figure 6.4 3rd eigenvalue stability region of negative real part within the parametric plane K p − K d
From the plots above, it becomes obvious that there is a region in the parametric space for
which the system can be stabilized. An important check for correctness is that the combination of the
three surfaces confirms that no stabilization is possible with only a P-control (i.e. K d = 0  K p  0 ),
neither for only a D-control (i.e. K p = 0  K d  0 ), nor for an open loop system without control (i.e.
K p = 0  K d = 0 ).
The boundaries stability region within the parametric space can be defined by illustrating the
intersection of the three surfaces with the 0 plane (Fig. 6.5).
70
FEEDBACK PD-CONTROL
Figure 6.5 Intersection pole surfaces with 0-plane. Limits of stability region
From this map, the stability region and stability thresholds can be defined as shown below
(Fig. 6.6). These stability limits can be defined by the inequality expression below. This directly
corresponds to the intersection of the real part of the eigenvalues with the zero-plane.
11867.65  K p  1.4496236 K d − 16785.332 (6.9)
Figure 6.6 Definition of the stability region
71
FEEDBACK PD-CONTROL
Moreover, the equilibrium point obtained for the PD-control nonlinear system is exactly the
same as the one found for P-control. This validates the conceptual frame mentioned in chapter 2.4,
for which the fixed point location is only dependent on the P-component. Whereas, the D-component
and the I-component define the type of fixed point.
Once a stability region has been identified, two analysis of this case study can be produced.
First, one can study the dynamic behaviour of the vehicle depending on the input control gain values
both within the phase portrait and for the evolution of the main variables of the system against time.
Second, one can draw the most general phase portrait of the system by running a set of simulations
with different initial conditions and perturbations, trying to picture out the tolerance in the variable
space for which the system is able to stay stable.
Along these lines, coming next several study cases are presented varying the control gain
parameters and the initial conditions.
6.1. SYSTEM RESPONSE STUDY BASED ON CONTROL

PARAMETERS
For this project, the control gain parameters have been defined to be constant in time. In this
manner, the control system is kept on a fundamental definition, avoiding more complex time
dependent tuning techniques or parametric space dependent tuning techniques, such as gain
scheduling.
Considering the first analysis type, the most simplified strategy is the definition of a fixed K d
value, defining a horizontal line through out the stability region plane presented in Fig. 6.6, and
present the results evolution for different K p values along this horizontal line. In this manner, the
effect K p and K d on the vehicle motion becomes more clear. The several case studies addressed in
this section are presented in Tab. 6.1.
Table 6.1 Summary of the study cases presented in this section
Gain parameters study cases based on K d = 28000

Case No Location Stable Region Kp value
Case 1 Close to Left Boundary In Stable Region 17000
Case 2 In the Middle of the Stable Region 20000
Case 3 Close to Right Boundary In Stable Region 23700
Case 4 Close to Left Boundary Outside Stable Region 16600
Case 5 In the Right Boundary In Stable Region 23804
Case 6 Close to Right Boundary Outside Stable Region 24200
Below, the behaviour of the system state-variables  z z& i is presented against time (Fig.
6.7, 6.9 and 6.11), proving that stabilization around a designed point occurs for all variables of
interest of the system and that the effect of the controller gain parameters on those variables is
globally very similar.
72
FEEDBACK PD-CONTROL
These plots feature not only the nonlinear simulation solution of the system state variables
 x1 x2 x5  =  z z& i (in black), but also the linearized solution derived analytically from the
linearization of the system around the fixed point (in blue) and the design equilibrium value (in red).
Using the strategy defined above, the case 1 featured results correspond the case for which
K p is significantly smaller than K d , located close to the left stability boundary K p = 16783.35 (Fig.
6.7). In this case, the input combination of gain parameters is K p = 17000 and K d = 28000 , whilst
the initial perturbation to the vehicles position is 1  10−4 m from the target air gap of the system.
Figure 6.7 Nonlinear and linearized solutions of the system against time for a K p value close to the left
boundary of the stability region
From Fig. 6.7, it becomes obvious that the system is stable around the equilibrium point
represented by the red line and the behaviour along time presented by all represented magnitudes is
very similar. An important conclusion drawn from this figure is that for cases where K d parameter
gain governs K p , i.e. is significantly larger than K p , the convergence rate is quite slow. This
translates in a longer time for the solution of variables, such as the vehicle position z and the
electrical current i , to reach values around the designed equilibrium value. This can be explained,
due to the fact that K p is one to one related with how fast the system approaches the equilibrium
solution by definition. Whereas the K d governs the error change amplitude. In this manner, a high
value of K d yield a very small amplitude of oscillation of the actual solution around the equilibrium
point, whilst it takes some time far all the variables to reach a band of values around the fixed point.
73
FEEDBACK PD-CONTROL
Besides, from these results, the relation between the vehicle vertical velocity and the vehicle
position becomes apparent. When the vehicle position oscillates with significant amplitude around
the system response value, the vertical velocity also oscillates around the equilibrium value as a
result of the relatively large position variations. Whereas, when the vehicle position oscillation
amplitude around a solution value is reduced to a small band and its response move closer to the
equilibrium value in a smooth trajectory, the velocity reduces its amplitude oscitation around the
design value drastically.
In this case, the nonlinear and the linearized solution are almost equal and no significant
offset can be observed, since the initial conditions are given close to the equilibrium point. The
observations described above also apply for the phase portrait representation of the system (Fig. 6.8).
In this map it becomes obvious the slow convergence rate. Nevertheless, the amplitude of the
trajectory oscillation around the fixed point is significantly small. In this plot, the initial condition is
marked with a circle and the direction is represented by an arrow.
Figure 6.8 Phase portrait of the system for a K p value close to the left boundary of the stability region
The second featured results correspond the case for which K p is about in the middle of the
stable region within both the right and the left boundaries defined in eq. 6.12. (Fig. 6.9). In this case,
the input combination of gain parameters is K p = 20000 and K d = 28000 , whilst the initial
perturbation to the vehicles position is 1  10−4 m from the target air gap of the system.
74
FEEDBACK PD-CONTROL
Figure 6.9 Nonlinear and linearized solutions of the system against time for a K p value centered within the
stability region
From this figure, it becomes obvious the effect that the shift of K p causes on the behaviour
of the solutions versus time, leading to a much faster convergence. This results from the fact that the
difference between K p and K d has been reduced and K d is no longer significantly larger than the
value of K p . As aforementioned, this follows directly from the fact that K p is one to one related
with how fast the system approaches the equilibrium solution by definition. Besides, the same
relation observed previously between the vertical velocity of the vehicle and the vehicle position is
observed in this case.
A good point about the results showcased above (Fig. 6.9) is that the shift on K p has not
been sufficiently large for the amplitude of oscillation of the actual solution around the equilibrium
point to change significantly, leading to a quite optimal behaviour of the derived solution, featuring a
fast convergence and a small error amplitude. Likewise the previous case, the nonlinear and the
linearized solution are almost equal and no significant offset can be observed.
The observations described above also apply for the phase portrait representation of the
system (Fig. 6.10). In this map it becomes obvious that the convergence is much faster than in the
previous case, drawing a progressive approach in form of a conical spiral, and the trajectory to the
stable equilibrium point is reached with a small amplitude of error oscillation about the point.
75
FEEDBACK PD-CONTROL
Figure 6.10 Phase portrait of the system for a K p value centered within the stability region
The third featured results correspond the case for which K p is close to the right boundary
K p = 1.4496236 K d − 16785.332 (Fig. 6.11). In this case, the input combination of gain parameters is
K p = 23700 and K d = 28000 , whilst the initial perturbation to the vehicles position is 1  10−4 m
from the target air gap of the system.
76
FEEDBACK PD-CONTROL
From these plots, an apparent change in behaviour with respect to time is observed. The trend
for all the magnitudes represented, i.e. system state variables, control variable and electromagnetic
force, is more similar for all variables. For all of them the convergence rate is very fast, in such a
way that the first step already overshoots and crosses the equilibrium point line. This means that
since the beginning of the solution simulation the variables are varying around the design point. This
results from the shift on K p coming closer to the value of K d . However, within the trade between
the effect of each of the gains to the overall system response, the fast convergence comes to a price,
since now the amplitude of the oscillation of the system response around the fixed point is much
larger and will keep on increasing as K p increases until the stability threshold is met.
The explanation for such a trade of can be derived not only through the natural definition of
each of the control components as previously done, but also through the system eigenvalues. In this
sense, within the stability region, at any point of this sub-space, the system is defined by one purely
real eigenvalue and two complex conjugate eigenvalues. Further proof of this is provided below, for
the explanation of the existence of a Hopf bifurcation in the system, as defined in chapter 2.4.2.
Nevertheless, for each combination of parameters within this sub-space the magnitude of both real
and negative parts changes.
Using the strategy defined above of fixing a K d value and defining a horizontal line across
the stability region, one can evaluate the change of the system eigenvalues along this horizontal line.
In this manner, it is observed that the main change on the eigenvalues’ nature between the left
boundary and the right boundary is translated in the magnitude of the real part of the complex
conjugate pair of eigenvalues. Along these lines, the magnitude of such real part is at its maximum
77
FEEDBACK PD-CONTROL
when the parameter combination is taken at the left boundary and it progressively reduces tending to
zero when moving to the right.
If one takes the parameter combination just at the right boundary, the real part of the complex
conjugate pair of eigenvalues is exactly zero. This is the so called bifurcation, the point in which
stability is lost and a change on the stability pattern of the system is defined. Now, if one approached
the limit of the real value of the complex conjugate poles through its negative value, one can see that
the stability pattern for a pair of complex conjugates with real part negative tending to 0 and non-
zero imaginary part corresponds to a stable centre. Relating this to the effect of the control gain
parameters tuning on the system response, this can be seen as the point at which K p takes over in the
aforementioned trade of and the amplitude on the error change around the equilibrium starts growing
exponentially, it is in this bifurcation point, when the fixed point will lose its stability and the
amplitude of this oscillations around the equilibrium will grow as a spiral trajectory on the phase
portrait plane move away from the fixed point will be observed.
Considering this explanation, one can understand that the phase portrait for the third set of
results presented below (Fig. 6.12) starts shaping up to a limit cycle.
In this case, since the amplitude of the oscillation of the system response around the fixed
point is much larger, the velocity solution cannot close in to the design value as in the previous case.
Moreover, in this case, some offset between the linearized and the nonlinear solutions starts to
appear (Fig. 6.11), which will accentuate when approaching the bifurcation point and will
completely differ right after it as showcased below.
Figure 6.12 Phase portrait of the system for a K p value close to the right boundary of the stability region
As aforementioned in this very same section 6 and in chapter 2.4, the existence of a Hopf
bifurcation is directly a consequence of finding a complex conjugate pair of eigenvalues with zero
78
FEEDBACK PD-CONTROL
real part. Therefore, an eigenvalue study is necessary. In this case, it is quite simple, since the system
is defined by three eigenvalues, which as performed before, can be plotted in a 3D space. A way of
showcasing the existence of this bifurcation, following the approach defined above on fixing K d and
varying K p , is to plot the analytical expression of the eigenvalues with respect to K p (Fig. 6.13). In
this case, the plot is produced fixing K d = 28000 .
Figure 6.13 Plot of the real part of the three eigenvalues of the linearized system with respect to K p
From the figure, the definition of the stability thresholds becomes evident and, consequently,
at the location where the pair of complex conjugates intersect the horizontal axis Re{s1,2,3 } = 0 and
the bifurcation is created. The red line overlaps with the green one, since the real part of the complex
conjugate pair of eigenvalues s2 and s3 is equal. Besides, one can conclude that no Hopf bifurcation
is expected around the left stability threshold, since in that case, the real part of the complex
conjugate pair of poles is not zero, but the purely real eigenvalue s1 becomes zero.
The behaviour of the system response around both boundaries for a fixed K d = 28000 is
presented below (Fig. 6.14 – 6.18). At the left boundary, the behaviour of the system for a K p value
a little bit lower than the boundary ( K p = 16600 ) is unstable. From the behaviour of the system state
variables and the control variable versus time (Fig. 6.14), it becomes obvious that the low value of
K p leads to a control that is not able to maintain the vehicle attracted at the target air gap level and it
falls downs, moving away from the design value. Given the correct definition of the system, it tries
to over compensate this by inputting more voltage to the system and raising the electrical current,
without success.
79
FEEDBACK PD-CONTROL
As a side note, from this figure (Fig. 6.14), it becomes obvious that the farther away from the
equilibrium point around which the system was linearized, the larger the divergence between the
linearized solution in blue and the nonlinear one in black.
Figure 6.14 Nonlinear and linearized solutions of the system against time for a K p value smaller than the
left boundary of the stability region
In this case, the phase portrait also shows an unstable response (Fig. 6.15). From the plot, one
can see that the system tries to oscillate around the fixed point in the initial stages, but it cannot hold
for long and starts deviating and moving away.
80
FEEDBACK PD-CONTROL
Figure 6.15 Phase portrait of the system for a K p value smaller than the left boundary of the stability region
Oppositely, if one focuses on the system response for a K p value exactly at the right stability
threshold ( K p = 23804 ) represented in Fig. 6.16 and 6.17, and a little larger than the right stability
boundary ( K p = 24200 ) represented in Fig. 6.18 and Fig. 6.19.
In this way, the creation of the Hopf bifurcation is captured in the system response. As
explained in chapter 2.4, the idea behind a Hopf bifurcation, is the change of stability pattern of the
system. In this case, fixing the K d parameter and varying the value of the K p , a critical value K p0 is
found, for which the stability of the system is changed from a point stability to a limit cycle. This is
captured by the nonlinear simulation. Nevertheless, the linearized system cannot capture this new
stability level and it translates into an unstable system response. Along these lines, for the nonlinear
solution, the fixed point becomes unstable and a spiral trajectory is drawn, moving away from the
point to a new stable level in the shape of an orbit. In this manner, the fact that the fixed point
derived in eqs. 6.4 – 6.6 loses its stability makes by definition the linearized system around this point
unstable. This becomes very obvious in the representation of the nonlinear and the linearized
solution for each state variable in Fig. 6.18, where while the nonlinear solutions in black stabilize at
a certain orbit the linearized solutions in blue grow exponentially to   .
Keep in mind, that this new stability level is not captured in the currently defined eigenvalue
problem, meaning that after K p values cross the stability boundary not all eigenvalues have a
negative real part. In this case, the pair of complex conjugate eigenvalues have positive real part.
This is explained by the fact that these eigenvalues are computed for the linearized system around
the fixed point, since, as aforementioned, the nonlinear system does not have eigenvalues. For this
reason, once the fixed point becomes unstable the eigenvalues also capture instability.
81
FEEDBACK PD-CONTROL
Figure 6.16 Nonlinear and linearized solutions of the system against time for a K p value exactly at the right
From this figure, it can be seen that the system is marginally stable, present a limit cycle
around the design equilibrium point with a certain amplitude. The linearized solution describes a
center stability pattern around the fixed point at a fixed amplitude, following the real parts of the
complex conjugate pair of eigenvalues tending through negative values to zero derived from the
linearized system. Such a limit cycle can be then presented in the phase portrait plane (Fig. 6.17).
Figure 6.17 Phase portrait of the system for a K p value exactly at the right boundary of the stability region
82
FEEDBACK PD-CONTROL
Once the input combination of gain parameters is defined crossing the right stability
boundary, the fixed point becomes unstable, the linearized solution for each of the state variables is
no longer stable, whilst the nonlinear system solution stabilizes at a new stability level, shaping an
orbit around the unstable fixed point (Fig. 6.18 and 6.19).
Figure 6.18 Nonlinear and linearized solutions of the system against time for a K p value after the right
Figure 6.19 Phase portrait of the system for a K p value after the right boundary of the stability region
83
FEEDBACK PD-CONTROL
6.2. SYSTEM RESPONSE STUDY BASED ON INITIAL CONDITIONS
Considering the second analysis type, the idea is to produce a significant amount of nonlinear
simulations with various initial conditions combinations, such that a variety of trajectories can be
plotted on the system phase portrait z − z& , providing a representation of the extend up to which the
system can stabilize the levitation system. With these results, one can prove that the system is stable
for the variable tolerances defined for the real design case with a fixed control gain parameter
combination.
For the sake of understanding, a significantly meaningful study case can consist on studying
the system response in time for four initial conditions combinations, one for each of the four
behavioural regions presented in Fig. 4.3, which were derived without any type of control. In this
manner, four cases are described in Tab. 6.2. All cases share the same electrical current, defined as
the design equilibrium point value x5s . Whereas the initial vehicle position and velocity is different
for each case, following Fig. 4.3.
Initial conditions study cases

Case No Region No in Fig. 4.3 x1 x2
Case 1 Region 1 20 0
Case 2 Region 2  0 (1 + 1e − 2) 0.03
Case 3 Region 3  0 (1 + 1e − 2) -0.03
Case 4 Region 4  0 (1 − 2e − 2) 0
These results are presented below (Fig. 6.20 – 6.26). For this analysis, the control gain
parameters have been fixed to K p = 20000 and K d = 28000 . As shown for the first type of analysis,
they are defined to be constant in time.
The first results feature the case of having an initial perturbation of the vehicle position of an
entire target air gap spacing 0.015 m from the equilibrium position (Fig. 6.20 and 6.21),
corresponding to region 1 of vehicle dynamics behaviour in Fig. 4.3.
84
FEEDBACK PD-CONTROL
Figure 6.20 Nonlinear and linearized solutions for study case within vehicle dynamics region 1
From this figure, the divergence between the linearized and the nonlinear solution for
locations far from the equilibrium point becomes obvious, although it is not very relevant in
magnitude and the quality of the stable outcome, it might be used to understand the difference in
behaviour between the two.
The behaviour of the three system state variables, the control variable and the
electromagnetic force show a stabilization of the system. Going somewhat more in depth, it is very
interesting to see how the system tries to react to a rather large perturbation of the vehicles levitation
air gap by increasing the electrical current. This is directly related with the definition of the
electromagnetic force leading to a trade off between air gap and electrical current to define an
electromagnet force that tends to move to a constant equilibrium value. This ultimately translates on
an increase on the input voltage required for the system to bring back the vehicle to the desired
design air gap.
The representation of the stabilization of the system from a distant initial point to the fixed
point, describing a stable spiral trajectory is captured in the phase portrait plane below (Fig. 6.21).
85
FEEDBACK PD-CONTROL
Figure 6.21 Phase portrait of the system for study case within vehicle dynamics region 1
The second results feature the case of having an initial perturbation of the vehicle position of
−2
1  10 m of the equilibrium position and an initial velocity of 0.03 m / s (Fig. 6.22 and 6.23),
corresponding to region 2 of vehicle dynamics behaviour in Fig. 4.3.
Figure 6.22 Nonlinear and linearized solutions for study case within vehicle dynamics region 2 and 3
86
FEEDBACK PD-CONTROL
In this case, the divergence between the nonlinear and the linearized solution is much smaller
since the initial position of the vehicle is much closer to the fixed points. From the plot of the
behaviour of the three system state variables, the control variable and the electromagnetic force
against time, it becomes clear that the stabilization of the system is much faster in this case than in
the previous. The behaviour of each of the represented magnitudes is very similar, implying that the
trade off between the variables, aforementioned for study case 1, is less relevant in this case to
compensate deviations from the equilibrium set up.
The representation of the stabilization of the system from case 2 initial conditions to the fixed
Oppositely to the second set, the third results feature the case of having an initial perturbation
of the vehicle position of 1  10−2 m of the equilibrium position and an initial velocity of −0.03 m / s
(Fig. 6.22 and 6.24), corresponding to region 3 of vehicle dynamics behaviour in Fig. 4.3.
In this case, the plot of the behaviour of the three system state variables, the control variable
and the electromagnetic force against time is very similar to the one for presented for the second set.
Therefore, the system can be already represented through Fig. 6.22 with respect to time. Then, the
representation of the stabilization of the system from case 3 initial conditions to the fixed point,
describing a stable spiral trajectory is captured in the phase portrait plane below (Fig. 6.24).
87
FEEDBACK PD-CONTROL
The fourth results feature the case of having an initial perturbation of the vehicle position of
−2
−2  10 m from the equilibrium position (Fig. 6.25 and 6.26), corresponding to region 4 of vehicle
dynamics behaviour in Fig. 4.3.
Figure 6.25 Nonlinear and linearized solutions for study case within vehicle dynamics region 4
The behaviour of the three system state variables, the control variable and the
electromagnetic force show a stabilization of the system. It is interesting to see how the system tries
88
FEEDBACK PD-CONTROL
to react to a negative perturbation of the vehicle levitation air gap, placing it closer to the rail, by
decreasing the electrical current, reducing the attractive action on the vehicle and making self wight
dominant on the force balance. This is directly related with the definition of the electromagnetic
force leading to a trade off between air gap and electrical current to define an electromagnet force
that tends to move to a constant equilibrium value.
The representation of the stabilization of the system from case 4 initial conditions to the fixed
From these results, one can conclude that the system has been stabilized for each of the
vehicle dynamics behavioural region studied previously for an open-loop system without control.
Nevertheless, further study is needed to understand the extend of these stabilization and prove that
the system guaranties a stabilization that covers the main state variables tolerances. A representative
way of showcasing such results is the plotting of the phase portrait of the system for different initial
conditions leading to different trajectories (Fig. 6.27 and 6.28).
Throughout the representation in the phase portrait plane z − z& of such a kind of studies, it is
found the existence of two stability patterns. Not only the systems presents a local point stability at
the previously derived fixed point, but also an outer unstable limit cycle exists (Fig. 6.27 and 6.28).
This outer unstable limit cycle is obtained by finding the limit initial conditions in the phase plane
space for which the system response does no longer converge to the fixed point. Consequently, a
stabilization region on the phase portrait plane is defined, meaning that the system is stable for any
initial conditions combination of ( z0 , z&0 ) within the space surrounded by the outer unstable orbit
and an initial electrical current fixed to the design equilibrium value i0 .
89
FEEDBACK PD-CONTROL
Moreover, the dimensions and the shape of the outer unstable orbit are dependent on the
control gain parameters choice ( K p , K d ) . The parameter combinations must be defined within the
K p − K d stable space are used. As a general rule of thumb, the larger the values of K p and K d , the
larger the stability space. In this report, two simple examples (Fig. 6.27 and 6.28) are presented for at
a local scale and for K p and K d values at close range to the ones used in other case studies.
The first example is produced for K p = 20000 and K d = 28000 (Fig. 6.27). A total of six
trajectories are represented in the plot, five of them stable, from which there is one trajectory for
each vehicle dynamics behavioural class from Fig. 4.3 and one starting at the equilibrium point, and
one of them unstable that allows to figure out how the unstable behaviour is drown. The plot is only
represented partly, focusing on a local range around the fixed point, but it is more extensive as z
increases. The unstable outer limit cycle is represented by a red line. In this case, since it was
obtained as a function, the whole cycle was not possible to be represented in one line, but two lines
were defined. One represents the top half and the other one the bottom half of the cycle. All the
initial conditions are listed in the legend and represented by a orange circle in Fig. 6.27 and 6.28.
listed in the legend and represented by a orange circle
From this figure, One can see that all trajectories starting inside the outer red unstable orbit
converge to the fixed point, i.e. the system is stable. Whilst, the trajectory for initial conditions given
outside it do not succeed to converge around the equilibrium point and ends up diverging and
moving towards the track outside the range presented in the graph. In this case, the stability region
in the phase plane is very small to the left of the equilibrium point. Whereas the range of the vehicle
vertical position rate is quite large.
90
FEEDBACK PD-CONTROL
To prove the effect that the tunning of K p and K d values on the shape of this orbit, a second
example is presented with larger control gain values that increases the stability for vehicle positions
closer to the track, but, as a trade off, the range of the vehicle vertical position rate is reduced. This
case study is produced for K p = 60000 and K d = 100000 (Fig. 6.28). As before, a total of six
trajectories are represented in the plot, five of them stable, from which there is one trajectory for
each vehicle dynamics behavioural class from Fig. 4.3 and one starting at the equilibrium point, and
one of them unstable that allows to figure out how the unstable behaviour is drown.
From Fig. 6.28, it becomes obvious that this second K p − K d combination results in a wider
stability region. This is specially useful to the left of the equilibrium point, since the stability space,
now, is much larger. This translates in the fact that, for such a control parameter combination, the
system is stable for a larger amounts of scenarios in which the levitation air gap is smaller than the
target value. Considering that the stability region in the parametric space (Fig. 6.6) has no upper
limit, it seems clear that larger gain parameter values with a similar ratio between each other provide
a larger limit cycle.
listed in the legend and represented by a orange circle
To sum up, this section shows that the stabilization of the system using a PD-control on the
voltage is possible and a large stable region is obtained on the parametric space K p − K d , allowing
91
FEEDBACK PD-CONTROL
the definition of several different behaviours of the system for a large range of control gain
parameters combinations.
Besides, the trade off in the effect that K p and K d generate in the system has been shown,
aiming at understanding better which is the suitable tunning of the control gains depending on the
system demands and constraint. Along these lines, the existence of a Hopf bifurcation in the system
linked to the choice of control parameters has been investigated. Finally, the region of stabilization
generated by the control has been explained and contrasted for different gains.
6.3. STUDY APPEARANCE OF A POSSIBLE NUMERICAL

ARTIFACT
To conclude this section of the report, a phenomenon about the stability of the system around
the fixed point around is addressed. Throughout the study of the EMS closed-loop model with a PD-
control, it has been shown that the nature of the stability of the equilibrium point derived by means
of the linearized system eigenvalue problem indicates that a point stability is expected at the design
equilibrium point. Nevertheless, the nonlinear solution differs a little at a very local scale around the
fixed points.
If one zooms in within the phase portrait plane around the fixed point, the appearance of a
small stable orbit around the equilibrium point becomes evident (Fig. 6.29). The order of magnitude
of the this orbit with respect to z is about rz = 10−8 and with respect to z& is about rV = 10−7 , which is
quite insignificant for the model outcome and it has no relevant effect on the system response. Since
this only appears in the nonlinear simulations and the radius of the orbit is sensitive to the ODE
solver used in the programming language, one may argue that this is solely due to numerical errors.
Figure 6.29 Small stable orbit around the fixed point in the phase portrait od the system
92
FEEDBACK PD-CONTROL
It is important to note that the system does not consider any type of damping on the interface
between the track, i.e. the support point, and the vehicle through the levitation electromagnetic field.
This is unrealistic, so a certain amount of artificial damping can be added to the system and, in case
this yields a significant reduction of this orbit or directly vanishes, it proves that the nature of such an
orbit is associated with numerical errors.
Following this approach, one can see a progressive regression of the radius of the orbit
already for the application of a 0.5% . For values of artificial damping around the 10% , the radius is
already moving to an order of magnitude of rz = 10−9 . However, it does not vanish for realistic
damping values. This investigation is inconclusive on whether the small limit cycles are of numerical
or physical origin. In case they are physically admissible, more advanced control definitions can be
applied, like linearized feedback techniques, time dependent control gain definitions or even gain
scheduling for any variable of interest.
As a side node, it is important to point out that since a successful stabilization of the system
has been reached already with a PD-control scheme, no I-component will be added for this problem
description avoiding adding more complexity to the system. This decision is taken considering the
fact that the integral component does not change the stability of the system but just smoothens the
system response [34].
93
DOUBLE CONTROL FEEDBACK PD-CONTROL AND FORCE SWITCH
7
ADDITION OF A FORCE
ON/OFF SWITCH IN THE
EMS SYSTEM WITH PD-
CONTROL
The previous section has proven a satisfactory stabilization of the system around the
equilibrium point and has featured the effect of the controller and the tunning of the contrail gains on
the system response. However, the implementation of such a control scheme has led to a side effect
that needs to be addressed and calibrated for a correct definition of the overall EMS levitation
system.
Throughout several different study cases presented above, it has been observed that for
unstable system set ups, or even close to unstable but still within the stabile region, the system tends
to shift the electrical current state variable solution to negative values. As explained in chapter 5, for
the choice of feasible fixed points of the systems, in this project, negative electrical current variables
are associated with a change of the polarization of the electromagnets in an attempt of the system to
stabilize the system and compensate the perturbations on the vehicle positions through the generation
of repulsive electromagnetic forces.
94
This is incompatible with the system definition provided in chapter 3, since this technological
scheme is thought to work only in attraction. In this manner, the definition of the electromagnetic
force defined for this project does not consider the possibility of becoming repulsive. This has been
guaranteed by the definition of the electromagnetic force as a relation between the squared electrical
current over the squared levitation air gap as shown in eq. 3.10. Consequently, the shift of electrical
current to negative values does not translate to the generation of repulsive electromagnetic forces,
but contributes to a build up of the electromagnetic force which grows in the opposite sense, i.e.
attractive, contributing to the contrary effect as the one the system intends.
This behaviour imposed by the control strategy is undesired, because in the response of the
system for rather small air gaps, lower than the design target one, the appearance of negative
electrical current trying to induce a change on the direction of the electromagnetic force ends up
leading to larger attractive electromagnetic force that pulls the system towards the track, making it
unstable. Hence, the representations of the stabilized region provided in chapter 6 (Fig, 6.27 and
6.28) can be improved by applying a complementary control strategy.
This problem can be solved in several different ways. One, that entails a significant addition
of complexity to the system, is the definition of time dependent control gain parameters and the
formulation of an optimization tuning scheme, which allows to define for each time step an optimal
combination of K p and K d that ensures stability of the system and always positive electrical current
values. Nevertheless, due to time constraints and following the aforementioned goal of keeping the
definition of the control fundamental, generic and simple, this option is not applied.
In this case, the solution that is applied is simpler. The main idea is that every time a negative
value of electrical current is computed a change of electromagnetic force is intended to be induced.
Thereby, it has been thought that the application of an on/off switch on the electromagnetic force
depending on the values of the electrical current could be applied. In this way the main condition of
operating the system only using attractive electromagnetic forces is met and, at the same time, the
appearance of unfeasible negative electrical current values is mitigated.
An on/off switch on the electromagnetic force can be implemented by defining an alternative

variable, the electrical current prediction i p which monitor the electrical current and predicts the next
step electrical current value. In this manner, the actual electrical current x5 can be defined as
x5 = i p  H (i p ) (7.1)
where H (i p ) is the Heaviside function on the predicted electrical current.
This allows to differentiate two system responses, as for x5  0 the electrical current is
defined to be zero x5 = 0 , whereas for x5  0 both variables are equal x5 = i p . Hence, for the first
case, the electromagnetic force is switch off and stops playing a role in the force balance of the
vehicle, allowing the self wight to take over and be the unique non-zero force acting on it. This is
then maintained until the moment in which the predicted electrical current crosses the zero axis again
and the electromagnetic force is switched on again.
95
In this manner, the vector of system state variables is defined as x

1 x2 
x5 p =
= z z& i p  and the system can be defined as follows
x&1 = x2
2
C x5 p
x&2 = g − sgn( x1 ) , if x5 p  0
m x12
(7.2)
x&2 = g , if x5 p  0
x1 x5
x&5 p = [ K p ( x1 −  0 ) + K d x2 + u0 − x5 p R − 2C 2p x2 ]
2C x1
The implementation of this control alternative does not change the system nature, meaning
that the fixed point and the eigenvalue problem are the same as the ones previously derive in chapter
6 for the PD-control.
The effect of this measure is represented below, initially featuring the system response
without the electromagnetic force switch control (Fig. 7.1) and, later, showcasing the system
response for the same system set up with the electromagnetic force switch control (Fig. 7.2). These
results are produced for a combination of K p = 20000 and K d = 28000 , and initial conditions
x1s = 0.0175 m and x2s = −0.43 m / s . Such an upwards initial vertical vehicle position rate is the
reason of negative electrical current values appearing in this case.
Figure 7.1 Nonlinear solution of the closed-loop EMS levitation system with PD-Control
96
From this plot it becomes obvious that the system tries to compensate a high upwards
velocity by using negative electrical current values in the initial time steps up to −10 A , whilst the
electromagnetic force is strictly positive defined. This contrasts with the figure below (Fig. 7.2), for
which the current becomes strictly positive defined, using the electromagnetic force on/off switch.
The electromagnetic force defined in the exact same way as before and the rest of state variables
present the same behaviour.
Figure 7.2 Nonlinear solution of the closed-loop EMS levitation system with PD-Control and force on/off
switch
Following the explanations and results above, it becomes obvious that the representation of
the stabilized regions previously derived (Fig. 6.27 and 6.28) need to be corrected by means of the
implementation of the on/off switch in the electromagnetic force. Along these lines, a new
representation of the stabilization region is produced using the same system set up as for Fig. 6.28,
for K p = 60000 and K d = 100000 (Fig. 7.3).
As before, a total of six trajectories are represented in the plot, five of them stable, from
which there is one trajectory for each vehicle dynamics behavioural class from Fig. 4.3 and one
starting at the equilibrium point, and one of them unstable that allows to figure out how the unstable
behaviour is drown. Likewise, the plot is only represented partly, focusing on a local range around
the fixed point, but it is more extensive as z increases.
From Fig. 7.3, it becomes obvious that the system stabilization region size is increased
significantly once the on/off switch on the electromagnetic force is applied, compensating for the
correction of the electrical current. At this point, the stabilization region obtained through this double
control scheme seems large enough to cover the main tolerances in levitation air gap and rate of
97
variation of the air gap for projected in the real project. Therefore, one can conclude that the
definition of such a control and the implementation of it in the EMS levitation system is satisfactory.
Figure 7.3 Phase portrait of the system for K p = 60000 and K d = 100000
98
CHAPTER 8: STABILITY STUDY VEHICLE VS TRACK INTERACTION WITH DOUBLE
CONTROL SCHEME OVER THE EMS
8
STUDY OF THE VEHICLE –
TRACK INTERACTION
The previous sections have served to define a control system based on a PD-control scheme
alongside an on/off switch on the electromagnetic force that has proven to be successful for the
stabilization of the inherently unstable EMS levitation system. Once the successful implementation
has been confirmed, its performance in the overall system considering the deflection of the beam
should be studied.
Thus, the main objective featured in this section is the study of the vehicle-track coupled
system and of the performance of the previously described control system in such a set up. Along
these lines, the assumption used in the derivation of the models presented in chapter 4, 5 and 6 on
considering the track equivalent beam to be infinitely stiff (eq. 4.1, 5.1 and 6.1) does no longer apply
to this system.
This study is based on the model presented in Fig. 3.7, which accounts for the modelling of
the first mode of the equivalent beam…
99
Hence, the vector of the system state variables is written as x 1 x2 x3 x4 

x5 p =

= z z& w w& i p  and the system of equations of motion can be defined as
x&1 = x2
2
C x5 p
x&2 = g − sgn( x1 − x3 ) , if x5 p  0
m ( x1 − x3 ) 2
x&2 = g , if x5 p  0
x&3 = x4 (8.1)
2
C x5 p
x&4 = g + sgn( x1 − x3 ) − 2 eqn x4 − n 2 x3 , if x5 p  0
M ( x1 − x3 ) 2
x&4 = g − 2 eqn x4 − n 2 x3 , if x5 p  0
x1 − x3 x5 p
x&5 = [ K p ( x1 −  0 ) + K d x2 + u0 − x5 p R − 2C ( x2 − x4 )]
2C ( x1 − x3 ) 2
Where n = keq / M is the natural frequency of the track beam,  eq = ceq / 2 keq M is the
damping ratio of the track beam
In this case, the equilibrium points have 5 components each, defining the system in a 5D
space, although the system is effectively a two degree of freedom system with a complementary
variable accounting for the electronics of the system. Thereby, this can be commonly referred to as a
one and a half degree of freedom system.
Nevertheless, one has to realize that the fact that five eigenvalues define the system, makes
the eigenvalue problem more challenging than for the previous study cases, since the real part on the
five eigenvalues can no longer be plotted as a 5D plot would be needed.
Thus, the equilibrium points of the system can be defined as ( x1* , x2* , x3* , x4* , x5 p * ) =
= {( x1 , x2 , x3 , x4 , x5 ) | x&1 = 0; x&2 = 0; x&3 = 0; x&4 = 0; x&5 = 0} and, as in previous study cases, not all
parameters in the system are positively defined.
In this case, the fixed points are exactly the same as the previously derived in chapter 5 and 6
with two extra components defining the beam deflection and the rate of deflection. Consequently, the
same exact procedure of analysing the feasibility of both alternatives is applied here, which yields
the same result. So the only admissible fixed point is the second one.
The inclusion of the lumped mass and the spring-dashpot system representing the beam only
introduces two new components/dimensions to the fixed point and significant complexity on the
symbolic description of components x1* and x5 p * . Despite of this, the fixed point still describes
exactly the design equilibrium position at the target levitation air gap, accounting for the deflection
of the track due to the vehicle hanging of it, with initial position rates zero for both masses (track and
100
vehicle) and with an initial electrical current corresponding to the initial voltage input on the system
to locate the vehicle at the design air gap.
( M + m) g ( M + m) g u
( x1* , x2* , x3* , x4* , x5 p * ) = ( 0 + , 0, , 0, 0 ) = (0.0164, 0, 0.0014, 0, 25.9257) (8.2)
K eq K eq R
Now, the eigenvalue problem is not only conducted as a function of K p , but also as a
function of K d since both are control gains, which are the tunable value that is assumed to make the
system stable. Therefore, one expects to be able to define different regions of stability depending on
the value of these parameters. However, now in a 5D space, which will not be possible to represent.
An alternative analysis will be proposed below.
Hence, the Jacobian matrix of the system and Jacobian evaluated at the fixed point are written below.
However, since the expressions get more and more involved as the complexity of the system
increase, several terms have been defined as complementary parameters in order to shorten the
expressions. These are  = (− R 2 gm + CK p 2 ) ,  = ( K p 0 − u0 ) ,  = CM 2 R 2 gm 4  2 ,
 = (−2CRgm − CK d u0 ) ,  = (− K d K p R 0 − R 2 gm + CK p 2 + K d Ru0 ) ,
 = (C 0 K d K p 3 +  K p 2 +  0 R 2 gmK d K p + 2 R 3 g 2 m 2 − gmK d u0 R 2 ) .
(8.3)
(8.4)
From eq. 8.5, the characteristic equation of the system can be expressed as
f ( ) =  5 + a1 4 + a2 3 + a3 2 + a4 + a4 (8.5)
Where each of the coefficients of this polynomials is derived from the Jacobian matrix
evaluated in the system fixed points by calculating det J |( x * , x * , x * , x * , x * ) − I and grouping the terms
1 2 3 4 5
corresponding to each of the polynomial terms order.
As addressed in chapter 2.4, according to Lienard-Chipart criterion, the necessary and

sufficient conditions for all eigenvalues of the characteristic equation eq. 8.6 to have a negative real
101
part are that the characteristic equation coefficients are all greater than zero and half of the Harwitz
determinant  i is grater than zero [9]. This translate in the following expressions
ai  0, i = 1...5
a1 1
2 = = a1a2 − a3  0 (8.6)
a3 a2
a1 1 0 0
a3 a2 a1 1
4 = = − a12 a4 2 − a1a2 2 a5 + a1a2 a3a4 + 2a1a4 a5 + a2 a3a5 − a32 a4 − a5 2  0
a5 a4 a3 a2
0 0 a5 a4
In this manner, the region of the parametric space K p − K d for which the system is stable
can be found (Fig. 8.1). This map has been produced by analysing the real part of the eigenvalues of
the system and its intersection with the zero plane.
Figure 8.1 Definition of the stability region
Then, the existence of a Hopf bifurcation with respect to K p , in the same way as the one
presented in chapter 6, can be proven using an algebraic criterion. This can be done if the following
theorems is met.
Theorem. The necessary and sufficient conditions for the characteristic equation eq. 8.6 to
have a pair of pure imaginary eigenvalue and the remaining n-2 eigenvalues to have negative real
parts are [9]
ai  0 (i = 1...5)  i  0 (i = n − 3, n − 5,...)   n −1 = 0 (8.7)
102
Applying the aforementioned theory and following Fig. 8.1 map of the stability region, the
following range of control parameters guaranties the stability of the region.
16783.35  K p  − 8715.1169 + 1.0133K d + 6.257  10−6 K d 2

(8.8)
K d  35177.95
Once a stability region has been identified, one can study the dynamic behaviour of the
vehicle depending on the input control gain values both within the phase portrait and for the
evolution of the main variables of the system against time.
The same strategy used in chapter 6 is applied in this case, fixing K d = 56000 and defining a
horizontal line through out the stability region plane presented in Fig. 8.1. Similarly, also several
different cases are defined depending on the value of K p . This can be found in Tab. 8.1.
Gain parameters study cases based on K d = 56000

Case No Location Stable Region Kp value
Case 1 Close to Left Boundary In Stable Region 20000
Case 2 In the Middle of the Stable Region 40000
Case 3 Close to Right Boundary In Stable Region 67600
Below, the behaviour of the three main system state-variables  z w i p  and the levitation
air gap  are presented against time (Fig. 8.2, 8.3 and 8.4), proving that stabilization around a
designed point occurs for all variables of interest of the system and that the effect of the controller
gain parameters on those variables is globally very similar.
These plots feature only the nonlinear simulation solution of the system state variables
x1 
x3 x5 p =  z w i p  and the levitation air gap  (in black), and the design equilibrium
value (in red).
Using the strategy defined above, the first featured results correspond the case for which
K p  K d , located close to the left stability boundary K p = 16783.35 (Fig. 8.1). In this case, the
input combination of gain parameters is K p = 20000 and K d = 56000 , whilst the initial perturbation
to the vehicles position is 1  10−4 m from the target air gap of the system.
103
Figure 8.2 Nonlinear solution of the system against time for a K p value close to the left boundary of the
stability region
From this figure, the same conclusions drawn in chapter 6 for the case study 1 are
extrapolated, amid the slow convergence rate with a very small amplitude of oscillation of the actual
solution around the equilibrium point. From Fig. 8.2, it is interesting to see how the levitation air gap
is defined as the relative difference between the track deflection and the vehicle displacement.
Nevertheless, one can conclude that no significant changes in the system response are observable,
due to the inclusion of the track motion.
The observations described above also apply for the phase portrait representation of the
system (Fig. 8.3) for the levitation air gap  − & . In this map it becomes obvious the slow
convergence rate . Nevertheless, the amplitude of the trajectory oscillation around the fixed point is
significantly small. It is important to note, that the phase portraits for such a case study, are not the
complete picture to the actual system behaviour, due to the simple reason that one is trying to
represent in a 2D plane the entire information of a 5D space. For this reason, unlike previous
sections, this chapter will not feature the phase portraits of the system, but only the solutions with
respect to time. Fig. 8.3 is the only exception on this rule, aiming at illustrate the aforementioned
phenomena with the perturbations on the trajectory, which unlike in previous sections does not draw
a perfect spiral around the design equilibrium point.
104
Figure 8.3 Phase portrait of the system  − & for a K p value close to the left boundary of the stability
region
The second featured results correspond the case for which K p is about in the middle of the
stable region within both the right and the left boundaries defined in eq. 8.9. (Fig. 8.4). In this case,
the input combination of gain parameters is K p = 40000 and K d = 56000 , whilst the initial
perturbation to the vehicles position is 1  10−4 m from the target air gap of the system.
Figure 8.4 Nonlinear solution of the system against time for a K p value centered within the stability region
105
extrapolated, amid the faster convergence rate due to the shift of K p with a little larger amplitude of
oscillation of the actual solution around the equilibrium point, compared to the previous one. From
Fig. 8.3, it is interesting to see how the levitation air gap is defined as the relative difference between
the track deflection and the vehicle displacement. Nevertheless, one can conclude that no significant
changes in the system response are observable, due to the inclusion of the track motion.
For this second case study, the interaction in motion between the rail track and the vehicle
becomes clearer that before. It is important to see how, once the vehicle is converging around the
design equilibrium point, the oscillation of the track deflection oscillation around the fixed becomes
much smaller. This is explained due to the fact that the vehicles motion oscillation has also reduced
and its influence on the variation of the deflection on the beam reduces drastically. Along these lines,
once this stabilization pattern occurs, at the same time the voltage input in the system and the
electromagnetic force also close in to the equilibrium value and damped out the oscillations around
it.
The third featured results correspond the case for which K p is close to the right boundary
K p = −8715.1169 + 1.0133K d + 6.257  10 −6 K d 2 (Fig. 8.5). In this case, the input combination of
gain parameters is K p = 67600 and K d = 56000 , whilst the initial perturbation to the vehicles
position is 1  10−4 m from the target air gap of the system.
extrapolated, amid an immediate convergence rate due to the shift of K p with a significantly larger
amplitude of oscillation of the actual solution around the equilibrium point, compared to all previous
106
cases. From Fig. 8.3, it is interesting to see how the levitation air gap is defined as the relative
difference between the track deflection and the vehicle displacement. Nevertheless, one can conclude
that no significant changes in the system response are observable, due to the inclusion of the track
motion. From this plot, it becomes obvious the presence of a Hopf bifurcation close to the parametric
combination chosen for this case study.
In this coupled vehicle-track model, there are five eigenvalues. In this case, there are two
pairs of complex conjugates and one real valued eigenvalue. As previously explained, the system is
stable if and only if the real part of both five is negative. This is indeed true within the stability
region presented in Fig. 8.1, as proven with the plot of the real part of all five eigenvalues against
K p , fixing a certain value of K d , which showcases the evolution of the eigenvalues within the
stability region explaining the behaviours presented in Fig. 8.2-8.5 and confirms the presence of a
Hopf bifurcation at the right boundary of such region (Fig. 8.6).
Figure 8.6 Plot of the real part of the three eigenvalues of the linearized system with respect to K p
Using the strategy defined above of fixing a K d value and defining a horizontal line across
the stability region, one can evaluate the change of the system eigenvalues along this horizontal line.
In this manner, it is observed that the main change on the eigenvalues’ nature between the left
boundary and the right boundary is translated in the magnitude of the real part of the complex
conjugate pair of eigenvalues. Along these lines, the magnitude of such real part is at its maximum
when the parameter combination is taken at the left boundary and it progressively reduces tending to
107
zero when moving to the right. Then, once these cross the right boundary of the stability region, their
real part becomes zero, implying the existence of a Hopf bifurcation.
Since the nature of the bifurcation is exactly the same as the one presented in chapter 6, no
further discussion is addressed on its shape and its development. For any doubt, see pages 77-83. It is
important to note, that the influence of the addition of the beam motion in the system, yield a larger
amplitude secondary limit cycle, after the subcritical Hopf bifurcation, making it a little more critical
than when only the vehicle is considered.
At the left boundary, the behaviour of the system for a K p value a little bit lower than the
boundary is unstable. From the behaviour of the system state variables and the control variable
versus time, it becomes obvious that the low value of K p leads to a control that is not able to
maintain the vehicle attracted at the target air gap level and it falls downs, moving away from the
design value. Given the correct definition of the system, it tries to over compensate this by inputting
more voltage to the system and raising the electrical current, creating an overshooting reaction that
crashes with the track.
In conclusion, likewise chapter 6, proving that the overall coupled vehicle-track system is
stable using a PD-control on the voltage and an on/off switch on the electromagnetic force and a
large stable region is defined on the parametric space K p − K d , allowing the definition of several
different behaviours of the system for a large range of control gain parameters combinations.
Moreover, one can conclude that the inclusion of the beam motion yields a little more
restrictive stable region with respect to K d , considering that there is a new bottom threshold in the
stability region at K d = 35177 , but is extended progressively with respect to K p , the higher the
value of K d . This becomes visually obvious in Fig. 8.1.
Besides, the the inclusion of the beam motions yield a little more subcritical Hopf
bifurcation. In this sense, the K p and K d trade off observed in chapter 6 is maintained in this
system.
Finally, this chapter also serves to feature the interaction between the motion of the track
modelled as a SDOF oscillator and the motion of the vehicle, and how its relative difference provides
the resulting levitation air gap, which shows a mixed behaviour resulting of the combination between
both motions.
108
CHAPTER 9: CONTROL DELAY SAFETY MARGIN FOR STABILITY
9
CONTROL DELAY SAFETY
MARGIN FOR STABILITY
This chapter serves to feature the definition of time delay in the control system and its effect
on the stability of the EMS levitation system.
Most dynamical systems have an inherent amount of delay and, if one builds a controller for
such a system, this must be taken into account. It is common to associate short delays with low
disturbances that can sometimes even be ignored, and long delays lead to larger disturbances that
may even trigger change in stability of the system. Along these lines, two main types of delay can be
defined. On the one hand, there are distorting delays for which the signal is distorted when its
original shape is altered in some way. This happens when time delay is different for various
frequencies that make up the signal (Fig. 9.1).
109
Figure 9.1 Example of distorting delay in time domain
Rather than thinking about these delays in the time domain, since each frequency is delayed
differently, it is more helpful to think about it in the frequency domain. How much each frequency is
delayed through a process can be represented in the phase portion of a bode plot, leading to the so
called phase delay.
On the other hand, there are disturbing delays for which the delay affects the whole signal
equally and results in a shift of the entire signal, which maintains the same shape but it is just
postponed by a certain amount of delay (Fig. 9.2). Such a type of delays is most commonly know as
transport delays.
Figure 9.2 Example of a transport delay in time domain
110
Both of these types of delays exist in real physical systems and they both can cause problems
in our controller design. This is because the controller has to use old information in order to
determine the current controller output, or it has to predict into the future how its output will impact
the system. Overall, this has the effect of effectively lowering the sample time of our system and,
therefore, to counter it one has to lower the bandwidth or speed of the controller. If the bandwidth is
not lowered, then the delay could cause stability issues. However, this slows down the system and
makes it less responsive.
If one knows the source of the delay, one can trade the cost of trying to remove it versus just
building around it. In order to know how to remove the delays, one needs to understand where they
come from in the system.
There are unintentional delays, which are a by-product of the design and not something that
was included on purpose. All real dynamic systems introduce phase delay or distort the system by
delaying some frequencies more than others. This is true for either mechanical or electrical. In
addition there are design components that might be added into the system for a necessary reason, but
then, increase phase delay as a by-product. These are thinks like low-pass filters to remove noise,
anti-aliasing filters prior to digitizing an analog signal, and building integrators into a controller.
This means that all sensors, actuators and processes create some amount of phase delay
across the spectrum, and depending on the controller design it can also add phase delay. If this is the
cause of the delays in our system, then, it might be necessary to look into faster sensors and actuators
that do not have as much lag.
However, phase delay is not the whole story, but transport delay needs also to be considered.
This one is associated to computation time, processing time, sampling time, slack time or even extra
safety margin times to account for unforeseen events.
There are also intentional delays. These are pauses that are designed into the system on
purpose, such as slack time, which can be added to the system accounting for the estimated
processing time plus an extra safety margin accounting for any random variations. This might be
preferable to occasionally having a process failed to complete within the deadline of a real-time
system.
The combination of all of these delays create the so-called dead time. That’s the non-
distorting transport delay part. If this dead-time is too long, one can find the system to become
unstable. This is our focus in this chapter, with the objective to determine a safety margin before the
system becomes unstable that could account for any processing, synchronization, sampling,
unforeseen cause, etc. delays.
In this case, our nonlinear system can be well approximated by a linearized model. For the
sake of simplicity, this chapter will be focused on the linearized system used in chapter 6, eq. 6.3,
rewritten below. Therefore, the initial model assumption of considering the track to be infinitely stiff
is applied. Therefore, the vector of state variables is defined as  x1 x2 x5  =  z z& i .
111
x&1 = x2
C x5 2
x&2 = g − sgn( x1 ) (9.1)
m x12
x1 x
x&5 = (u − x5 R − 2C 52 x2 )
2C x1
However, now, the definition of the controlled voltage across the electromagnet coil present
in the 3rd equation is different. Aiming at accounting for a generic extra delay within the definition of
the controller, a variation of time is added for those variables dependent implicitly on time within the
error-based PD-control loop.
u (t − t ) = K p [ x1 (t − t ) −  0 ] + K d x2 (t − t ) + u0 (9.2)
The implementation of the time delay presented in eq. 9.2 is chosen to be conducted in the
linearized system for simplicity. The fixed point is defined in the same way as in chapter 5 and 6, as
shown in eq. 9.3. Then the Jacobian of the eq. 9.1 system evaluated at the fixed point reads
u0
( x12* , x22* , x52* ) = ( 0 , 0, ) (9.3)
R
 
 
 
 0 1 0 

 
 2C 2 2C 
J |( x * , x * , x * ) =  0 −  (9.4)
 m 3 m 2 

1 2 5
 
 −
 
 2C 2 2C 
Where  = (CK p + R gmC ) ,  = ( Rgm + K p gmC ) ,  = (− R 2 gm + CK p 2 ) ,

 = (− RK d K p 0 − R 2 gm + CK p 2 + RK d uo ) ,  = ( K p 0 − u0 ) ,  = (−2CRgm − CK d u0 ) .
Then, the same method used in chapters 4, 5 and 6 can be used to reduce the two first rows of
the matrix system to an equivalently single equation with variable x1 and x5 , and coefficients
variables x1 and x5 . In this step, the time delay is applied, only to the variables involved in the
control.
x1 (t ) − J 21 x1 (t ) + J 23 x5 (t ) = 0
&&
(9.5)
x&5 (t ) − J 31 x1 (t − t ) − J 32 x&1 (t − t ) + J 33 x5 (t ) = 0
Similarly, no conclusions can be drawn about stability of the system due to the presence of
K p and K d and the eigenvalue problem needs to be derived. This is done in the Laplace domain.
112
Note that the Laplace transform for a force with time delay is defined as
 [ f (t − t )] = F ( s) exp(− st ) .
&&x1 (t ) − J 21 x1 (t ) + J 23 x5 (t ) = 0 ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
Laplace trans .
→ s 2 xˆ1 − J 21 xˆ1 + J 23 xˆ5 = 0
(9.6)
x&5 (t ) − J 31 x1 (t − t ) − J 32 x&1 (t − t ) + J 33 x5 (t ) =  (t ) sxˆ5 − J 31 xˆ1 exp( −st ) − J 32 sxˆ1 exp(−st ) + J 33 xˆ5 = 1
1 + J 31 xˆ1 exp(− st ) + J 32 sxˆ1 exp(− st ) 1 + J 31 xˆ1 exp(− st ) + J 32 sxˆ1 exp(−st )
xˆ5 =  s 2 xˆ1 − J 21 xˆ1 + J 23 ( ) = 0→
s + J 33 s + J 33
J 23
−
s + J 33 1
xˆ1 = =− (9.7)
J 23 ( J 31 + J 32 s) exp(− st ) 1 3
s − J 21 +
2
[ s + J 33 s + ( J 23 J 32 exp(− st ) − J 21 ) s − J 21 J 33 + J 23 J 31 exp(− st )]
2
s + J 33 J 23
1 3
⎯
⎯⎯ ⎯⎯⎯⎯
Poles
→ [ s + J 33 s 2 + ( J 23 J 32 exp(− st ) − J 21 ) s − J 21 J 33 + J 23 J 31 exp(− st )] = 0
J 23
The equation defining the poles is no longer polynomial due to the presence of exponential
functions with the complex variable s in the argument. Therefore, to determine the poles of such an
equation, it is chosen to approximate the exponential function through the first two terms in the
Taylor’s expansion exp(− st )  1 − st + higher order ( s ) . Hence, the polynomial equation reads
1442443
0
1 3
[ s + J 33 s 2 + ( J 23 J 32 (1 − st ) − J 21 ) s − J 21 J 33 + J 23 J 31 (1 − st )] = 0 (9.8)
J 23
Hence, the eigenvalues can be plotted with respect to t , which allows to see for which
values of t the system is stable and for which not. Considering the same strategy followed in
previous chapters of fixing a value of K d and varying the value of K p within the stable region in the
parametric space, three cases have been differentiated. The first one corresponds to the case for
which K p is close to the left boundary (Fig. 9.3).
113
Figure 9.3 Eigenvalues plotted versus the time delay for K p value close to the left boundary of the stable
region
From Fig. 9.3, it becomes obvious that the system behaviour is exactly the same as described
when plotting the eigenvalues against K p for the PD-control Fig. 6.13. For such a value of K p the
real-valued eigenvalue is close to cross the real axis, which means that the system is close to the
stability boundary. Whilst the pair of complex conjugate eigenvalues have a rather large negative
real part. For K p − K d combinations close to the left boundary of stability provide the largest time
delay safety margin.
The second one corresponds to the case for which K p is in the middle of the stability region
defined in the parametric space (Fig. 9.4).
114
Figure 9.4 Eigenvalues plotted versus the time delay for K p value in the middle of the stable region
As before, the system behaviour is confirmed as both eigenvalues start from the negative
complex half-space with time delay zero. For such a K p − K d combination the reduction of the
safety margin is reduced significantly.
The third one corresponds to the case for which K p is close to the right boundary of the
stability region defined in the parametric space (Fig. 9.5).
Figure 9.5 Eigenvalues plotted versus the time delay for K p value close to the right boundary of the stable
region
115
As before, the system behaviour is confirmed as both eigenvalues start from the negative
complex half-space with time delay zero, but now the pair of complex conjugates real part is much
closer to zero, i.e. to the bifurcation. For such a K p − K d combination the reduction of the safety
margin is to its minimum.
Once the safety margin with respect to the eigenvalues has been acknowledge, this allows the
designer of the error-based PD-control to account for the software processing delay, the signal
sampling delay, the actuators delay and some slack time for unforeseen events, making sure that the
system is still stable.
To sum up, this section has shown that enough time delay safety margin can be obtained.
This directly depends on the K p − K d combination and, it has been showcased that this safety
margin maximizes when K p is close to the left boundary of the stable region.
116
CHAPTER 10: CONCLUSIONS AND RECOMMENDATIONS
10
CONCLUSIONS AND
RECOMMENDATIONS
This chapter serves to conclude the project summarizing all the results, relating them to the
original problem statement and thesis objectives, and providing several recommendations and
alternative research lines for any student interested in continuing to study this topic.
10.1. CONCLUSIONS
The main goals of this thesis can be summarized in three main points. First, we aimed at
investigating dynamic behaviour of the EMS levitation system without controller. Second, we
focused on investigating the implementation of an error-based PD-control in the levitation system.
Third, we attempt to studying the vehicle-structure interaction through the EMS system after the
control implementation.
The EMS levitation system has been confirmed to be unstable by nature, meeting Earnshaw’s
theorem, and the vehicle dynamics have been understood as a trade off between electromagnetic
force and gravitational forces. Such a trade off is governed by the initial value problem for levitation
air gap magnitude and rate of variation of the gap.
117
The P-control has proven not sufficient to stabilize the system, although it provides already a
certain level of response trying to counteract the loss of attractive force for positions far from the
track. Oppositely, for the PD-control a stable region defined for different values of the control gain
parameters can be obtained. This system has proven to be stable for a significant range of initial
conditions air gap and air gap variation rate.
Besides, the existence of a subcritical Hopf bifurcation has been confirmed for a critical
proportional gain parameter K p value laying on top of a boundary of the stability region and a given
fixed K d value. In this sense, it has been confirmed the direct relation between the value of the
proportional gain parameter K p and the convergence rate of the system response to the equilibrium
point. Whilst a direct relation between the value of the derivative gain parameter K d and the error
variation amplitude around the design equilibrium value was derived as well. In this manner, a trade
off has been defined, setting the basis for a tuning optimization of the parameters.
The implementation of an on/off switch on the electromagnetic force has proven to improve
the system and avoid unrealistic negative electrical current. This turns out to increase the range of
initial conditions, i.e. perturbations, in air gap and air gap variation rate, for which the system can be
stabilized. Such a range has been proven to meet the main tolerances in the system state variables.
For a more complex control strategy, the implementation of this switch can be avoided by tunning
the control gains accordingly.
The overall vehicle-track system has been proven to stabilize by means of the error-based
closed-loop PD-control system. While the on/off switch on the electromagnetic force improves the
system response by avoiding the appearance of unrealistic negative electrical currents and enlarges
the stable region in the phase portrait space. A stable region with respect to the control gain
parameters has been derived using the eigenvalues of the linearized system.
Moreover, the inclusion of the beam motion yields a little more restrictive stable region with
respect to K d , considering that there is an additional boundary at the bottom side of the stability
region. This stable region is extended progressively with respect to K p , the higher the value of K d
becomes. In this case, the existence of a subcritical Hopf bifurcation has also been confirmed for this
model, for which the secondary stability pattern has a little larger amplitude, making it more critical.
The effect of the control gains parameter set up on the system has been analysed, confirming the
same trade off observed in previous studies.
Finally, the safety margin for a certain amount of time delay on the control system response
has been derived, providing a range of delay magnitude to be accounted when choosing a certain
model of measuring device, processor, etc.
10.2. RECOMMENDATIONS
For further studies, it would be highly advised to implement certain optimization algorithms
to tune the control parameters, such that the most efficient control system response is derived,
avoiding the need of secondary mechanisms such as the on/off switch on the electromagnetic force.
118
The two degrees of freedom system is a simplification and accounting for the actual
continuous system is important, making special emphasise in the vehicle-structure interaction. In this
sense, this project proposed the modelling of the track-beam as an Euler Bernoulli beam, for
simplification and following the assumptions listed in chapter 3. However, considering that recent
design updates propose shorter sections and that important shear strains, like the ovalization of the
tube, can be relevant in the track structural behaviour. For this reason, future investigations could
shift the modelling of the track infrastructure to the Timoshenko beam theory.
Besides, for simplification a track span section is modelled as a simply supported beam.
Nevertheless, this is not fully correct. The track-beam can be modelled in a more correct way by
considering a continuous beam. One of the simplest starting points is the consideration of an infinite
continuous beam laying on a continuous homogeneous viscoelastic support, accounting for a
homogeneous bed. The vehicle can then be added using either a moving load or an equivalent spring
dashpot oscillator system accounting for the electromagnetic levitation force between the two bodies.
For such a longitudinal beam analysis, if a continuous beam is introduced, there are more
chances to account for the role of intermediate supports. Along these lines, the periodicity of the
supports could be introduced. This can be addressed by considering an infinite beam defined on a
periodically inhomogeneous support, accounting for the periodicity of the supports. As before, the
vehicle can then be added using either a moving load or an equivalent spring dashpot oscillator
system accounting for the electromagnetic levitation force between the two bodies.
Moreover, irregularities present on the track can be added to the levitation gap  model as a
time and longitudinal coordinate dependent function p( x, t ) , so as to account for the unrealistic
assumption of a perfectly smooth track. For an advanced modeling, this term could be represented by
a Power Spectral Density function (PSD) accounting for the random nature of such irregularities.
Other simplified and less realistic options are using a time dependent term with sinusoidal shape
p(t ) = Ap sin(2 Vm / wLt ) at the specific cross-section cut being studied.
If there is interest in studying the cross-sectional study further, the horizontal guidance
system could be added, so as to understand the interaction between the vertical levitation system and
the horizontal one. Such a horizontal guidance distance is only activated when facing bendings, track
switches and track irregularities. For this reason, the mechanics of the system must be defined in a
significantly different way, as gravitational forces do not play a role, but centrifugal/centripetal pair
of forces and lateral pulse forces.
119
BIBLIOGRAPHY
BIBLIOGRAPHY
[1] A. Schafer and D. G. Victor, "The future of mobility of the world population,"
Transportation Research Part A - Policy and Practice, vol. 34, p. 171, 2000.
[2] P. Vallance and J. Norman, "The Future of Mobility: A time of unprecedented change in
the transport system," Governmetn Office for Science; Foresight, UK, January 2019.
[3] H. Ritchie, "Our World in Data - Climate change and flying> what share of global CO2
emissions come from aviation?," University of Oxford, 22nd October 2020. [Online].
Available: https://ourworldindata.org/co2-emissions-from-aviation. [Accessed 13th
April 2022].
[4] B. Dodson, "Beyond the hype of Hyperloop: An analysis of Elon Musk's proposed
transit system," 22nd August 2013. [Online]. Available: https://newatlas.com/hyperloop-
musk-analysis/28672/. [Accessed 1st March 2022].
[5] S. Earnshaw, "On the nature of molecular forces which regulate the constitution of the
Luminiferous Ether.," in Transcription of the Cambridge Philosophical Society,
Cambridge, M. A. of St. John's Colledge, 1842, p. 97.
[6] X. Chen, W. Ma and S. Luo, "Study on stability and bifurcation of electromagnet-track
beam coupling system for EMS maglev vehicle," Nonlinear Dynamics, vol. 101, pp.
2181-2193, 2020.
[7] X. Chen, W. Ma, S. Luo and R. Zou, "A vehicle-track beam matching index on EMS
Maglev transportation system," Arch. Appl. Mech., 2019.
[8] H. S. Han, B. H. Yim, N. J. Leec, Y. C. Hur and S. S. Kim, "Effect of the guideway's
vibrational characteristics on the dynamics of a Maglev vehicle," Vehicle System
Dynamics, vol. 47, no. 3, pp. 309-324, 2009.
[9] J. Hu, W. Ma, X. Chen and S. Luo, "Levitation Stability and Hopf Bifurcation of EMS
Maglev Trains," Mathematical Problems in Engineering, 2020.
[10] J. H. Li, J. Li and D. F. Zhou, "Self-excited vibration problems of Maglev vehicle-bridge
interaction system," J. Cent. South Univ., vol. 21, no. 11, pp. 4184-4192, 2014.
[11] X. Li, D. Wang, D. Liu, L. Xin and X. Zhang, "Dynamic analysis of the interactions
between a low-to-medium speed Maglev train and a bridge: field test results of two
typical bridges," Proceedings of the Institution of Mechanical Engineers, Part F:
Journal of Rail and Rapid Transit, vol. 36, pp. 2039-2059, 2018.
[12] X. H. Shi, L. H. She and W. S. Chang, "The bifurcation analysis of the EMS maglev
vehicle-coupled-guideway system," Acta Mechanica Sinica, vol. 36, pp. 634-640, 2004.
[13] R. P. Talikdar and S. Talukdar, "Dynamic Analysis of High-Speed MAGLEV Vehicle-
Guideway," Urban Rail Transport, vol. 2, no. 2, pp. 71-84, 2016.
[14] J. Xu, C. Chen, D. Gao, S. Luo and Q. Qian, "Nonlinear dynamic analysis on Maglev
train system with flexible guideway and double time-delay feedback control," Journal of
Vibroengineering, vol. 19, no. 8, pp. 6346-6362, 2017.
[15] J. D. Yau, "Vibration Control of Maglev vehicles travelling over a flexible guideway,"
Journal of Sound and Vibration, vol. 321, pp. 184-200, 2009.
[16] M. Zhang, S. Luo and C. Gao, "Research on the mechanism of a newly developed
levitation frame with mid-set air spring," Vehicle System Dynamics , vol. 56, no. 12, pp.
1797-1816, 2018.
120
BIBLIOGRAPHY
[17] T. E. Alberts, G. Oleszczuk and A. M. Hanasoge, "Stable levitation control of

magnetically suspended vehicles with structural flexibility," in American Control
Conference, 2008.
[18] A. M. G. Jacob and N. M. V. Monteiro, "A new concept of superelevation in magnetic
levitation - prodynamic," Transportation Systems and Technology, vol. 4, no. 4, pp. 77-
111, 2018.
[19] "Maglev Trains - McGill," 2007. [Online]. Available:
https://www.cs.mcgill.ca/~rwest/wikispeedia/wpcd/wp/m/Maglev_train.htm. [Accessed
28th February 2022].
[20] Z. Liu, Z. L. Li and X. Li, Maglev Trains: Key Underlying Technologies, Berlin:
Springer, 2015.
[21] "Technology Development - Hardt Hyperloop," 2020. [Online]. Available:
https://hardt.global/technology-development/. [Accessed 2nd March 2022].
[22] Cassandra, "How does Hyperloop work? Everything you need to know about magnetic
levitation - alphr," 16th November 2020. [Online]. Available:
https://www.alphr.com/technology/1006815/how-hyperloop-works-launch-magnetic-
levitation/. [Accessed 2nd March 2022].
[23] "XP-2 Test Vehicle - Virgin Hyperloop," 2022. [Online]. Available:
https://virginhyperloop.com/pegasus?a=tech. [Accessed 2nd March 2022].
[24] J. L. He, D. M. Rote and H. T. Coffey, "Study of Japanese electrodynamic-suspension
Maglev systems," Technical Report, 1994.
[25] "Magnetic Levitation - UCSB Physics," [Online]. Available:
https://web.physics.ucsb.edu/~lecturedemonstrations/Composer/Pages/68.78.html.
[Accessed 2nd March 2022].
[26] S. Yadav, S. Verma and S. Nagar, "Optimized PID Controller for Magnetic Levitation
System," International Federation of Automatic Control, vol. 49, no. 1, pp. 778-782,
2016.
[27] R. Goodall, "Dynamics and control requirements for EMS Maglev suspensions," in
Proc. Int. Conf. Maglev, Shanghai, Octrober 2004.
[28] E. Kong, J.-S. Song, B.-B. Kang and S. Na, "Dynamic reponse and robust control of
coupled Maglev vehicle and guideway system," Journal of Sound and Vibrations, vol.
330, pp. 6237-6253, 2011.
[29] H.-S. Han and D.-S. Kim, Magnetic Levitation: Maglev Technology and Applications,
Dordrecht: Springer, 2016.
[30] K. Wang, S. Luo and W. Ma, "Dynamic characteristics analysis for a new-type maglev
vehicle," Adv. Mech. Eng., vol. 9, no. 12, pp. 1-10, 2017.
[31] A. V. Metrikine, Dynamics, Slender Structures and an Introduction to Continuum
Mechanics, Delft University of Technology: Lecture notes CT4145, 2020.
[32] E. M. J. Vicca, "Instability of a high-speed vehicle moving through a tunnel embedded
in a viscco-elastic half-plane," Delft University of Technology , 2019.
[33] S. H. Strogatz, Nonlinear Dynamics and Chaos with Applications to Physiscs, Biology,
Chemistry and Engineering, Boulder, CO: Westview Press, 2015.
[34] P. Wolf and e. al., "P, I, D, PI, PD, and PID Control," University of Michigan, December
2021.
[35] O.-o. c. system, "X-Engineer," [Online]. Available: https://x-engineer.org/on-off-
control-system/. [Accessed 7th April 2022].
121
BIBLIOGRAPHY
[36] A. Urquizo. [Online]. Available:

https://en.wikipedia.org/wiki/PID_controller#/media/File:PID_en.svg. [Accessed 7th
April 2022].
[37] K. Cui and X. Qin, "Virtual reality research of the dynamic characteristics of the soft soil
under metro vibration loads based on BP neural networks," Neural Computing and
Applications, pp. 1-10, 2017.
[38] F. Orban, "Damping of materials and members in structures," in 5th International
Workshop on Multi-Rate Processes and Hysteresis, Vancouver, 2011.
[39] D. Scott and J. Free, "Maglev: How they're getting trains off the ground?," Popular
Science, p. 135, 1973.
122
ANNEX A: ELECTROMAGNETISM
ANNEX A:
ELECTROMAGNETISM
The phenomenon of magnetic levitation is developed through magnetic fields between

magnetic objects. Such magnetic fields are generated by the movement of electric charges. Let 𝑞 be
⃗⃗⃗, as shown below (Fig. A.1) [29].
a moving charge at a velocity 𝑣⃗ that produces a magnetic field 𝑩
Then, this magnetic field is defined as:
r r
ur  qv  (r / r )
B= 0
(10.1)
4 r2
From eq. 10.1, the magnitude of the field is given by
0 q v sin 
B= (10.2)
4 r2
123
Figure A.1 Magnetic field produced by a moving electric charge [29]
This principle is the basis of permanent magnets and solenoids to produce magnetic fields.
Such magnetic fields are generally represented by field lines with no beginning or end, forming
closed loops. The field is spatial, as shown in Fig. A.2. The field lines come from the North pole (N)
and enter the South pole (S). The lines are continuous and do not cross each other [29].
Figure A.2 Magnetic fields in solenoid (left) and permanent magnet (right) [29]
An attractive force is generated between two magnetized objects when the field lines go from
one to another (Fig. A.3. a). Whereas if the same poles are facing each other, the field lines are
pressured, resulting in a repulsive force between them (Fig. A.3. b) [29].
124
Figure A.3 Representation attraction (a) and repulsion (b) forces [29]
Although the permanent magnets and solenoids are the ones responsible for generating the
magnetic field, some materials can be magnetized by an external magnetic field. This leads to the so-
called magnetic substance. The most common types of these materials are iron, nickel, cobalt and
most of their alloys. Ferromagnetic materials have the strongest capacity for magnetism. Whereas,
diamagnetic materials do not respond to an applied magnetic field [29]. This phenomenon will take
part in our system for the magnetization of the rail leading to a magnetic track.
A flux 𝜙 is a series of field lines. The value of the flux corresponds to the number of field
lines. The more lines the stronger the magnetic field and the larger the flux. Consequently, the flux
value is also seen as a measure of the magnetic field’s strength. Its unit is the weber [Wb]. 1 Wb
corresponds to 108 field lines. Then, the flux density 𝐵 is defined as the flux per unit area normal to
a magnetic field. Its unit is the tesla [T] [29]. 𝐵 can be defined as follows

B= [Wb / m 2 ] (10.3)
A
If electrons flow through a conductor, a magnetic field is produced around it. This field is
called an electromagnetic field, and the properties of the field are the same as the properties of a
permanent magnet. The direction of the electromagnetic field is perpendicular to the wire and moves
in the direction the fingers of your right hand would curl if you wrapped them around the wire with
your thumb in the direction of the current, as shown in the following illustration (Fig A.4) [29].
125
Figure A.4 Magnetic field formed around a conductor carrying current [29]
Magnetic permeability  represents the relative ease of establishing a magnetic field in a

given material. The permeability of free space is called 0 , and its value is 0 = 4  10−7 [ H / m] .
Relative permeability of any material r =  / 0 compared to 0 is a convenient way to compare its
magnetization. For steels, the relative permeabilities range from 2000 to 6000 or higher. Thus, if an
iron core is wound by coils carrying currents, almost all of the flux produced by the coils goes
through the iron core, not air, which has a smaller permeability than that of iron [29].
Reluctance is a magnetic resistance in materials, which is the counterpart of electrical

resistance. Let l be the length and A the area of the flux path [29]. Then reluctance  is defined as
l
= (10.4)
A
Then, analogously to the voltage or electromotive force, the magnetomotive force is the cause
of the magnetic flux in a magnetic circuit. Therefore, it can be defined as the effective current flow
applied to the core. This can be mathematically described by eq. 10.5 as a function of the electrical
current I and the number of coils turns N and graphically illustrated by the magnetic circuit below
(Fig. A.5) [29].
Fm = NI [ A  N o turns] (10.5)
126
Figure A.5 Simple magnetic circuit [29]
Then, all the magnetic fields produced by the current will remain inside the core because the
core’s permeability is higher than air and the flux of the magnetic circuit is defined as [29]
Fm
= (10.6)

An electromagnet is defined as a type of magnet in which the magnetic field is produced by

an electric current. Generally, it consists of a large number of closely spaced turns of wire that create
the magnetic field. The wire turns are often wound around a magnetic core made from ferromagnetic
materials. The magnetic core concentrates the magnetic flux and makes a more powerful magnet.
The main advantage of an electromagnet over a permanent magnet is that the magnetic field can be
quickly changed by controlling the amount of electric current in the winding. The direction of a
magnetic field is dependent on the direction of the electric current [29].
The degree to which a magnetic field by a current can magnetize a material is called
magnetizing force H , and it is defined as the magnetomotive force Fm per unit length of material
[29]. That is
Fm NI
H= = [ A  no turns / m] (10.7)
l l
H is not related to a material’s property. The magnetic flux B induced in the material
depends upon the nature of the material [29], and the relationship between H and B is defined by
B = H (10.8)
Note that the flux in the materials is related linearly to the applied magnetomotive force in the
unsaturated region, and approaches a constant value regardless of the magnetomotive force in the
saturated region. Due to this behaviour, the operational region of EMS systems should be located in
the unsaturated region [29].
127
Applying an alternating current to the windings on the core instead of a direct current with a
frequency, the flux in the core traces out a path abcdeb in Fig. A.6. This is because the amount of
flux present in the core depends not only on the amount of current applied to the winding of the core
but also on the previous history of the flux in the core. This dependence on the preceding flux history
and the resulting failure to trace flux paths is called hysteresis. The path bcdeb traced out in Fig. A.6
as the applied current changes are called a hysteresis loop. If the frequency of an alternating current
is changed, the path is also changed with a different residual flux. This property may lower the
control performance at higher frequencies in a levitation system with electromagnets [29].
Figure A.6 The hysteresis loop traced out by the flux in a core when the alternating current is applied to it: a)
alternating current, and b) hysteresis loop [29]
When a conductor is exposed to a time-varying magnetic field, a voltage is induced across it,
as shown in the figure below (Fig. A.7). This can be mathematically described using Faraday’s law
of induction. Electric generators and motors as well as magnetic levitation systems are based on this
electromagnetic induction. The polarity of induced voltage depends on the direction of relative
motion [29]. Expressed in the form of an equation for induced voltage eind
d
eind = − N (10.9)
dt
Where the minus sign in eq. 2.9 is an expression of Lenz’s law [29].
128
Figure A.7 Electromagnetic induction [29]
If a conductor has electrical resistance, an electrical current flows in the conductor. This
current is called induced current iind . This is represented in the following figure (Fig. A.8) [29].
Figure A.8 Induced current [29]
If the directions of the flux lines from magnets and conductors are the same, the flux density
increases. In contrast, if the directions are opposite, the flux density decreases. The resulting forces
129
are exerted towards a weak magnetic pressure region from a stronger pressure region, as illustrated
below (Fig. A.9). This is the operating principle of electric motors [29].
Figure A.9 Forces on current-carrying conductors in magnetic field [29]
In electromagnetism and electronics, inductance is the property of a conductor by which a

change in current flowing through it induces a voltage or electromotive force in both the conductor
itself (i.e. self-inductance) and any nearby conductors (i.e. mutual inductance). A changing electric
current through a circuit that contains inductance induces a proportional voltage that opposes the
change in current, as illustrated below (Fig. A.10). It is customary to use the symbol L for
inductance. The unit for inductance is the henry [ H ] . The relationship among the parameters for a
coil with inductance L is defined as [29]
dI
V = IR − L (10.10)
dt
where
N 2 A
L=
l
Eq. 10.10 indicates that inductance opposes the applied voltage. This property to oppose
building up currents influences the control performance of levitation systems with electromagnets
[29]. This is the theoretical derivation of the core part of the equation of motion relating the electrical
variables to the mechanical variables governing the EMS system defined for this project. This
relation is effectively derived using Kirchhoff’s total voltage law.
Figure A.10 Coil's reaction to increasing current [29]
130
ANNEX B: STRUCTURAL DYNAMICS REVIEW
ANNEX B:
STRUCTURAL DYNAMICS
REVIEW
B.1. SINGLE DEGREE OF FREEDOM

Let us take a simple Single Degree Of Freedom (SDOF) mass-spring-dashpot system like the
one represented below (Fig. B.1), where M is the lumped mass, k0 is the spring stiffness, c0 is the
damping coefficient, F (t ) is a random time-dependent force acting on the lumped mass, and y (t ) is
the DOF representing the vertical motion of the lumped mass.
Figure B.1 SDOF mass-spring-dashpot system with F(t) random time-dependent external load
131
The equation of motion for such a system can be then obtained using the Newton’s second
law, expressing the equilibrium of forces acting on the lumped mass. For y (t ) = y notation, the
equation is expressed as
F(t ) + Fspring (t ) + Fdashpot (t ) = M &y& (11.1)
Where Fspring (t ) is the force exerted by the spring such that Fspring (t ) = −k0 y and Fdashpot (t ) is
the viscous resistance associated to the dashpot such that Fdashpot (t ) = −c0 y& .
Keep in mind that all terms in eq. 11.1 are vectors. Substituting the aforementioned force
definitions in eq. 11.1 and projecting the result onto the y-axis, the following expression is obtained
F (t )
M && y + 2n y& + n 2 y =
y + c0 y& + k0 y = F (t ) →&& (11.2)
M
Where  is a measure for the viscous damping in the system such that  = (1/ 2n )(c0 / M ) ,
and n is the natural frequency of the undamped system such that n = k0 / M .
For which non-zero initial conditions can be defined as
y (0) = y0 ; y& (0) = v0 (11.3)
At this point, two main cases can be differentiated: free vibrations and forced vibrations. The
only main difference is the right-hand side term of the equality presented in eq. 11.2.
B.1.1. Free Vibrations
This is the case in which the external force F (t ) is absent. Then eq 11.2 reduces to a
homogeneous differential equation
y + 2n y& + n 2 y = 0
&& (11.4)
This equation governs small free vibrations of the mass-spring-dashpot system [31]. The
general solution of this second-order differential equation can be written as
y (t ) = Yn exp( sn t )
2
(11.5)
n =1
Then, if eq. 11.5 is substituted in eq. 11.4 the following expression representing the
characteristic equation of the system is obtained
sn 2 + 2n sn + n 2 = 0 (11.6)
Where sn are the characteristic exponents, also called eigenvalues of the system, which can
be easily obtained by solving the second-order polynomial characteristic equation eq. 11.6.
132
s1,2 = −n  n  2 − 1 (11.7)
Thereby, the general solution for a single degree of freedom mass-spring-dashpot can be
described.
y (t ) = Y1 exp( s1t ) + Y2 exp( s2t ) = exp( −nt )[Y1 exp(n  2 − 1 t ) + Y2 exp(n  2 − 1 t )] (11.8)
Where Y1 and Y2 are arbitrary constants [31].
Four different solutions/behaviours can be distinguished depending on the value of the

damping coefficient  :
1)  = 0 → Undamped vibrations:
The characteristics exponents, in this case, are defined as s1,2 =  in =  i k0 / M . Hence, the
general solution is expressed as
y (t ) = Y1 exp(in t ) + Y2 exp(−in t ) (11.9)
Then, Euler’s formula can be used to express the exponential function as a

trigonometric relation, which yields eq. 11.10, where A and B are unknown real-valued
constants.
y (t ) = Y1 (cos(nt ) + i sin(nt )) + Y2 (cos(nt ) − i sin(nt )) =

(11.10)
= (Y1 + Y2 ) cos(nt ) + i (Y1 − Y2 ) sin(nt ) = A cos(nt ) + B sin(nt )
A and B can be found by substituting eq. 11.10 into the initial conditions eq. 11.3.
A = y0 ; Bn = v0 (11.11)
Thereby, the equation of motion of the undamped lumped mass M within the mass-
spring- dashpot system from its equilibrium point is expressed as
v0
y (t ) = y0 cos(n t ) +
sin(nt ) (11.12)
n
For a better comprehension of the results plotted in the figure below (Fig. B.2), eq.
11.12 can be equivalently written as
y (t ) = A0 cos(1t − 0 ) (11.13)
Where the amplitude A0 and the phase angle 0 read
133
v0 v0
A0 = y0 2 + ( ) 2 ; 0 = arctan( ) (11.14)
n n y0
By plotting the displacement-time curve in accordance with eq. 2.24 (Fig. 2.16), one can see
that vibrations of the mass-spring system are perfectly sinusoidal and last forever. In reality
there almost always exist some damping (energy loss) in vibration systems. This damping
causes the oscillatory motion induced by the initial disturbance to be reduced to zero over
time [31].
yo
Figure B.2 Free Vibrations of the mass-spring system  = 0 [31]
2)   −1    1 → Super-critically damped vibrations:
In this case, both characteristic exponents are real-valued. Hence, the general solution is
expressed as
y (t ) = A exp[(−n − n  2 − 1)t ] + B exp[(−n + n  2 − 1)t ] (11.15)
Where coefficients A and B can be determined by imposing initial conditions. In this case,
let’s consider the particular case y (0) = y0 ; y& (0) = 0 and substituting these into eq. 11.15
yields
y0 (−n − n  2 − 1) y0 (−n + n  2 − 1)
A=− ; B= (11.16)
2n  2 − 1 2n  2 − 1
Thereby, the equation of motion of the super-critically damped lumped mass M within the
mass-spring-dashpot system from its equilibrium point is expressed as
134
y0
y (t ) = {(−n + n  2 − 1) exp[(−n − n  2 − 1)t ] +
2n  − 1 2
(11.17)
+(−n − n  2 − 1) exp[(−n + n  2 − 1)t ]}
In connection with this solution, it should be noted that both s1 and s2 are negative, s2
having a bigger absolute value. Thus, the displacement y of the mass has the same sign as y0
and approaches zero as a limit when time tends to infinity. The displacement-time diagram
plotted in accordance with eq. 11.17 is shown below (Fig. B.3), where we see that the motion
is not a vibration at all, but an aperiodic motion in which the suspended mass, after its initial
displacement, gradually creeps back toward the equilibrium position but takes theoretically
infinite time to get there [31].
yo
Figure B.3 Aperiodic free motion of mass-spring-dashpot system   −1    1 [31]
3) −1    1    0 → Sub-critically damped vibrations:
In this case, both characteristic exponents are complex-valued. Aiming at clearly showcase
the physical significance, a change of the eigenvalues structure introducing real positive
values 1 = n 1 −  2 is helpful. Then the characteristic exponents read
s1,2 = −n  i1 (11.18)
Hence, the general solution is expressed as
y (t ) = exp(−nt )[ A cos(1t ) + B sin(1t )] (11.19)
As performed in previous cases, the eq. 11.19 must be substituted in the initial conditions, in
this case general initial conditions eq. 11.3, so as to obtain the expressions for arbitrary
constants A and B .
135
v0 n y0
A = y0 ; B = + (11.20)
1 1
Thereby, the equation of motion of the sub-critically damped lumped mass M within the
mass-spring-dashpot system from its equilibrium point is expressed as
v0 n y0
y (t ) = exp(−nt )[ y0 cos(1t ) + ( + ) sin(1t )] (11.21)
1 1
For a better comprehension of the results plotted in the figure below (Fig. B.4), eq. 11.21 can
be equivalently written as
y (t ) = A0 exp(−nt ) cos(1t − 0 ) (11.22)
Where the amplitude A0 and the phase angle 0 read
v0 n y0 2 v + n y0
A0 = y0 2 + ( + ) ; 0 = arctan( 0 ) (11.23)
1 1 1 y0
By plotting the displacement-time curve in accordance with eq. 11.22 (Fig. B3), one can see
that this motion is vibratory in nature and represents damped free vibrations. The main effect
of damping on free vibrations is that it dissipates energy from the system leading to a decay
of the amplitude of vibrations [31].
y
yo
Figure B.4 Free vibrations of mass-spring-dashpot system −1    1    0 [31]
In this manner, Fig. B.4 shows that each time that cos(1t − 0 ) becomes equal to  1 , the
time-displacement curve is tangent to one of the envelopes  A0 exp(−nt ) at the ordinates
t = 0 / 1 , t = (0 +  ) / 1 , t = (0 + 2 ) / 1 ... . This quantity is called the amplitude of
vibration. It is seen that, owing to damping, the amplitude gradually diminishes with
time and that the rate of decay depends on the damping factor ratio  . The time T1 = 2 / 1
136
required to complete one cycle of the motion is called the period of vibration and its
reciprocal f1 = 1 / 2 is called frequency of vibration [31].
4)  = 1 → Critically damped system:
This is a special case of aperiodic motion for which the value of the damping coefficient
corresponds to the so called critical damping, which is defined as
c 0 = 2n m = 2 km (11.24)
Along these lines, the characteristic exponents of the system are equal s1,2 = −n , i.e. there is
one eigenvalue with multiplicity 2. This yields the following general solution
y (t ) = A exp(−nt ) + Bt exp(−nt ) (11.25)
B.1.2. Forced Vibrations
This is the case in which the external force F (t ) is present. Then, the equation of motion of
the system is described by the second order inhomogeneous differential equation eq 11.2. Thereby,
one should address this problem by breaking the solution into two pieces: a homogeneous solution
and a particular solution. The homogeneous solution yh (t ) is exactly the same as the aforementioned
general solution for the free vibrations motion. Whereas the particular solution y p (t ) carries with it a
bit more complexity.
In case the shape of the function of time F (t ) is well known, as in the case of a harmonic
force, generally, it can be obtained choosing a solution defined with a generic function with
unknown parameters with the same shape as the load, i.e. with equal time signature than the load.
Then, one can compute the unknown coefficients substituting the generic solution function selected
into the equation of motion of the system. Lastly, the final solution to the problem is obtained by
summing both parts as
y (t ) = yh (t ) + y p (t ) (11.26)
As an example, consider the dynamic response of the mass-spring-dashpot system to a

harmonic sinusoidal force given by F (t ) = F0 cos(t ) . Then the equation of motion of the system
considering the same notation used in the section above reads
y + 2n y& + n y = f 0 cos(t )

&& (11.27)
Where f 0 = F0 / M .
The harmonic solution (i.e. general solution) to eq. 11.27 is given by eq. 11.19. Whereas, the
particular solution to eq. 11.27 is found as
137
y p = Yc cos(t ) + Ys sin(t ) (11.28)
Where Yc and Ys are unknown constants which can be found by substituting eq. 2.39 into eq.
11.27. yielding
(− 2Yc + 2nYs + n 2Yc − f 0 ) cos(t ) + (− 2Ys − 2nYc + n 2Ys )sin(t ) = 0 (11.29)
This equation can be satisfied for all values of t only if the expressions in parenthesis vanish.
Thus, for calculating Yc and Ys , one has to solve the following system of linear algebraic equations
[31]
− 2Yc + 2Ys + n 2Yc = f 0

(11.30)
− 2Ys − 2Yc + n 2Ys = 0
From which it is obtained

(n 2 −  2 )
Yc = f 0
(n 2 −  2 ) 2 + 4 2n 2 2
(11.31)
2n
Ys = f 0
(n −  2 ) 2 + 4 2n 2 2
2
The last step, is to introduce the initial conditions in the overall solution so as to find the
initial value problem.
However, the approach on this particular solution changes if the load function shape is not
well known or it is wanted to be kept generic, as in the case of forced vibrations under a general
disturbing force. Let’s now consider the following equation of motion and initial conditions
y + 2n y& + n 2 y = f (t )
&&
(11.32)
y (0) = y0 ; y& (0) = v0
Assuming that the external force per unit mass f (t ) is an arbitrary function of time. For any
instant of time one elementary impulse f (t ')dt ' can be considered. In accordance with Newton’s
second law, this one impulse imparts to each unit of mass an instantaneous increase in velocity
dx& = f (t ')dt ' . Treating the increment of velocity as if it was an initial velocity at instant t ' and using
eq. 11.21, it can be concluded that the corresponding displacement of the mass-spring-dashpot
system at any later time will be [31]
f (t ')dt '
y (t ) = exp(−n (t − t ')) sin(1 (t − t ')) (11.33)
1
Note that this expression holds both for sub- and super-critically dumped vibrations, but only
the nature of 1 changes.
138
Since each impulse f (t ')dt ' between t ' = 0 and t ' = t has a like effect, as a result of the
continuous action of the external force, the following displacement of the mass is obtained [31]
1 t
y (t ) =
1

0
f (t ') exp( −n (t − t '))) sin(1 (t − t ')) dt ' (11.34)
The complete solution to the initial value problem eq. 11.32, eq. 11.21 and 11.34 need to be
added up. Bear in mind that eq. 2.32 represents the system response to the initial conditions [31]
v0 n y0
y (t ) = exp(−nt )[ y0 cos(1t ) + ( + ) sin(1t )] +
1 1
(11.35)
1 t
+  f (t ') exp(−n (t − t '))) sin(1 (t − t '))dt '
1 0
An important remark is the concept of steady state response. Which is defined as a stationary
situation that remains after the transient motion has died out over time [32].
ysteady (t ) = lim y (t ) (11.36)

t→ 
In case the system is dynamically unstable, the vibrations doe not decay, but grow exponentially,
which means that the steady state response is never reached [32].
B.1.3. SDOF Stability with respect to Damping
On a side note, aiming at understanding a bit more the role of damping in the system’s
stability, certain definitions about the stability of the system can be listed. For a simple SDOF system
as the one presented in this section, the stability of the system is directly related with damping.
In a dynamical system, one can define instability as the phenomenon that takes place when
small initial perturbation grows in time. In this case, the system does not return to its equilibrium
position. This phenomena cannot happen for the critically damped vibration and the undamped
vibrations case (i.e.  = 0   = 1 ), since they are always stable. However, when   0    1
instability may occur [32].
On the one hand, if the system presents sub-critically damped vibrations, the term directly
determining the stability of the system for both general solution forms eq. 11.19 and eq. 11.22 is the
term exp(−nt ) . Whereas the rest of the equation describes the sinusoidal vibration without any
grow or decay. Remind that, for this specific case, both characteristic exponents have a positive real
part. In this manner one can argue that: if 0    1 , the exponent remains negative, leading to an
exponential decay which describes stability, i.e. fading of small perturbation induced vibrations over
time; if −1    0 , the exponent becomes positive, leading to an exponential growth which
describes instability, i.e. small perturbation induced vibrations grow over time.
On the other hand, if the system presents super-critically damped vibrations, the relation is
directly derived, considering the aperiodic motion nature. If   1 , the system is stable as negative
139
exponentials governs the general solution expression. If   −1 , the system is unstable as positive
exponentials governs the general solution expression.
Superposing the aforementioned main points, it is concluded that the SDOF system is
unstable for negative damping coefficient values   0 .This means that, when the system is
negatively damped, the characteristic exponents have negative real part.
B.2. MULTIPLE DEGREE OF FREEDOM

In this thesis several models present a higher level of complexity, entailing a discretization
into more than one single degree of freedom. In these cases, the principles discussed above for SDOF
systems also applies. However, the equation of motion of the system is defined in matrix form.
 M  &&y + C y& + K  y =  f  (11.37)
Where  M  , C and K  are N  N squared matrices accounting for mass, viscous
damping and stiffness, respectively. Where N is the number of degrees of freedom of the system.
Through this matrices, the coupling between several degrees of freedom is defined. Several different
coupling situations can be observed depending on the nature of the system. If a spring or a spring-
dashpot connects two lumped masses, the coupling between the degrees of freedom associated to
such lumped masses is then manifested in the stiffness matrix and, if applicable, the damping matrix.
However, for other type of connections between two lumped masses, the coupling between the
degrees of freedom associated to such lumped masses may appear in the mass matrix, yielding an
inertial coupling. Even in some cases, several different degrees of freedom might be decoupled.
If there is no damping, the solution of the system can be assumed to have the form below [29]
 x =  X  exp(isnt ) (11.38)
Substituting this general solution in the undamped system equation of motion yields
 X  exp(isnt )( M sn  +  K ) = 0

2
(11.39)
The system has a non-trivial solution if the determinant of its coefficient matrix vanishes
[31]. This defines the eigenvalue problem. The solution to this problem should yield N eigenvalues
and when the eigenvalues are substituted back into the original set of equations, the values of  x that
correspond to each eigenvalue are eigenvectors. These eigenvectors represent the mode shapes of the
systems [29]. This means that it is the shape that the system takes when excited to the
eigenfrequency associated to the eigenvector. This is important in order to understand the interaction
between the degrees of freedom.
B.3. SIMPLY SUPPORTED EULER-BERNOULLI BEAM MODAL

VIBRATION
140
As already mentioned in chapter 1, one of the objectives of this thesis is to analyse and
understand the dynamics, the coupling and the stability vehicle-track infrastructure. Along these
lines, as most of the sources presented in the literature review (chapter 2.1.5) pointed out, the
flexibility of the guideway has a significant impact on the levitation stability ( [30], [16]).
In this project, focus has been put on the tubular infrastructure of the Hyperloop design. This
is one of the main differences from Maglev applications. Indeed one of the innovations of this thesis
is the focus of the structural behaviour of this feature and its effect on the system stability.
Along these lines, one span of track is modeled as a simply supported, evenly elastic with
equal cross-section Euler-Bernoulli beam, which yields the following equation
 4 w( x, t )  5 w( x, t )  2 w( x, t )
EI +c + A = f ( x, t ) (11.40)
 x4  x4  t t2
Where w( x, t ) is the transverse deflection as a function of the longitudinal axis of the beam x
and time t , E is the Young’s modulus, I is the second moment of inertia of the beam cross-section,
c is the structural element damping,  is the mass density of the beam material, A is the cross-
sectional area of the beam, and f ( x, t ) is the external vertical loading on the beam.
Simply supported boundary conditions for a beam of length L supported at each end can be
described as
 2 w( x, t )
x = 0 : w( x, t ) = =0
 x2
(11.41)
 2 w( x, t )
x = L : w( x, t ) = =0
 x2
B.4. LAPLACE TRANSFORM
The Laplace transform is an integral transform which serves to transform signals and
functions from the time domain to the frequency domain. This tool is particularly useful in order to
solve linear ordinary differential equations, since in the frequency domain, these become simply
algebraic equations. The Laplace transform of a function G(t ) is defined as
+
t [G (t )] = g ( s ) =  G (t ) exp(− st )dt (11.42)

0
Where G(t ) is a function of the real variable t and s is the complex variable, i.e. s =  + i .
G(t ) is called the original function and g ( s) is called the image function [31].
The operations, which are valid for the Laplace transform are shown in Table B.1.
141
Table B.1 Operations for Laplace transform [31]
This method will prove to be very useful to compute the poles of our linearized system
around the fixed points, given that the equations of motion are all linear ordinary differential
equations. Besides, it makes the implementation of a control within the system easier to analyse and
formulate in the frequency domain in case a study of the transfer functions, the frequency bandwidth
of stabilization or a phase delay analysis wants to be performed.
Below, a table of Laplace transforms is presented (Table B.2), for the formulation of general
functions of the real variable t and the complex variable s .
Table B.2 Table of Laplace transforms [31]
142
143
ANNEX C: STABILITY
ANNEX C:
STABILITY
C.1. ONE-DIMENSIONAL FLOW BIFURCATIONS
The most basic bifurcation type is the saddle-node bifurcation, which is the basic mechanism by
which fixed points are created and destroyed. As a parameter is varied, two fixed points move toward
each other, collide and mutually annihilate. The prototypical example of a saddle-node bifurcation is
given by the first-order system
x& = r + x 2 (12.1)
Where r is a parameter, which may be positive, negative or zero. When r is negative, there
are two fixed points, one stable and one unstable (Fig. C.1. a). As r approaches zero from below, the
parabola moves up and the two fixed points move towards each other. When r = 0 , the fixed points
coalesce into a half-stable fixed point at x* = 0 (Fig. C.1. b). This type of fixed point is extremely
delicate, as it vanishes as soon as r  0 and, afterwards, there are no fixed points at all (Fig. C.1. c).
In this case, bifurcation occurs at r = 0 [33].
Figure C.1 Saddle-node bifurcation example
144
ANNEX C: STABILITY
Therefore, if one represents the saddle-node bifurcation as a continuous stack of vector fields
a picture of the generation of the bifurcation can be illustrated (Fig. C.2), which then can be
represented in continuous lines accounting for stability and instability (Fig. C.3).
Figure C.0.2 Vector field stack [33] Figure C.0.3 Saddle-node bifurcation [33]
This example is the so called normal form of the saddle-node bifurcation as the dynamics
typically look like eq. 12.1, so it is kind of the prototypical example of this pattern [33].
There are certain cases where fixed points must exist for all values of a parameter and can
never be destroyed. Such a fixed point may change its stability as the parameter is varied. The
transcritical bifurcation is the standard mechanism for such changes in stability (Fig. C.4). The
normal form for a transcritical bifurcation is [33]
x& = rx − x 2 (12.2)
Figure C.4 Transcritical bifurcation example
Another kind of bifurcation is the so called pitchfork bifurcation, which is common for
physical problems that have symmetry. Such a bifurcation can be supercritical or subcritical. The
supercritical pitchfork bifurcation normal form is
x& = rx − x3 (12.3)
When r < 0, the origin is the only fixed point, and it is stable. When r = 0, the origin is still
stable, but much more weakly so, since the linearization vanishes. Now solutions no longer decay
exponentially fast, but instead the decay is a much slower algebraic function of time (Fig. C.5).This
lethargic decay is called critical slowing down in the physics literature. Finally, when r > 0, the
145
ANNEX C: STABILITY
origin has become unstable. Two new stable fixed points appear on either side of the origin,
symmetrically located at x* =  r (Fig. C.5 and C.6) [33].
Figure C.5 Supercritical pitchfork bifurcation vector field [33]
Figure C.6 Supercritical pitchfork bifurcation [33]
Whilst for the supercritical case, the cubic term is stabilizing, for the subcritical case, the
cubic term is destabilizing. Which translates in the opposite pattern (Fig. C.7).
Figure C.7 Subcritical pitchfork bifurcation [33]
146

MSC Report JoanMasSoldevilla 4947096 FinalVersion

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MSC Report JoanMasSoldevilla 4947096 FinalVersion

Uploaded by

Copyright:

Available Formats

J.

DYNAMIC ANALYSIS AND STABILITY STUDY OF THE

in partial fulfilment of the requirements for the degree of

at the Delft University of Technology,

Student number: 4947096

An electronic version of this thesis is available at http://repository.tudelft.nl/.

1.2. PROBLEM STATEMENT

Unlike other Hyperloop designs, Hardt Hyperloop’s proposal is based on an electromagnetic

The design and implementation of electromagnetic suspension systems in Hyperloop entail

1.3. RESEARCH OBJECTIVES

1. Extrapolate EMS system definition for the Hardt Hyperloop design.

1.a) Define the formulation and the critical parameters involved.

2. Design an error-based closed-loop control following a fundamental approach that can

1.4. REPORT OUTLINE

2.1. MAGNETIC LEVITATION TECHNIQUES

Figure 2.1 Representation of the different magnetic levitation systems [18]

2.1.1. Electromagnetic suspension (EMS)

Figure 2.2 EMS system used in the Transrapid [18]

2.1.2. Electrodynamic suspension (EDS)

2.1.3. Comparison of different magnetic levitation technologies

Each of the aforementioned systems has advantages and disadvantages. A good

2.1.4. Stability of magnetic levitation technologies

2.1.5. Research and major achievements in the field

2.2. ELECTROMAGNETICS FUNDAMENTALS

The phenomenon of magnetic levitation is developed through magnetic fields between

An electromagnet is defined as a type of magnet in which the magnetic field is produced by

In electromagnetism and electronics, inductance is the property of a conductor by which a

Considering the aforementioned definitions of electromagnets, it is important to introduce the

2.3. STRUCTURAL DYNAMICS REVIEW

t [G (t )] = g ( s ) =  G (t ) exp(− st )dt (2.3)

2.4.1. One dimensional flow

The one-dimensional flow can be represented by a dynamic system described as follows

Where x(t ) is a real-valued function of time t and f ( x) is a smooth real-valued function of

Figure 2.5 Phase portrait of the system [33]

Nevertheless, so far, only graphical methods have been discussed to determine in a

Afterwards, Taylor’s expansion can be applied in eq. 2.5.

f ( x* +  ) = f ( x* ) +  f '( x* ) + O( 2 ) (2.6)

& =  f '( x* ) + O( 2 ) (2.7)

&   f '( x* ) (2.8)

Thus, if a fixed point is regarded as an equilibrium solution, the approach to equilibrium is

Figure 2.6 Circle flow [33]

2.4.2. Two-dimensional flows

or a nonlinear system is considered and then the linearized approximation as described in

x(t ) = exp(t ) v (2.11)

the characteristic equation becomes

Expanding the determinant yields

Where  = trace( A) = a + b and  = det( A) = ad − bc .Then, the solution of the quadratic

The aforementioned statements can be proven considering the eigenvalues definition

In this manner, assuming   0 provides different eigenvalues (i.e. zero-multiplicity). Then,

Figure 2.10 Example star node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 ≠ 0 [33]

Figure 2.11 Example degenerated node for eigenvalues 𝜆1 = 𝜆2 = 𝜆 = 0 [33]

Figure 2.13 Diagram representing the classification of fixed points [33]

The stability of nodes and spirals is determined by 𝜏:

▪ For 𝜏 < 0 ⇔ 𝑅𝑒(𝜆1 ) = 𝑅𝑒(𝜆2 ) < 0 → Stable fixed points

Consequently, 𝑥 ∗ is asymptotically stable if it is both attracting and Liapunov stable.

Figure 2.14 Representation of attracting point and Liapunov stability [33]

Figure 2.15 Limit cycle classification [33]

Figure 2.16 Parameter family of closed orbits

Figure 2.19 Closed-loop feedback control general structure [35]

Figure 2.20 PID Controller closed-loop feedback scheme [36]

PD-control is combination of feedforward and feedback control, because it operates on both

Proportional-integral-derivative control is a combination of all three types of control

3.1. RAIL INFRASTRUCTURE MODEL